talk-data.com

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

Welcome to the Best of ICCV series, your virtual pass to some of the groundbreaking research, insights, and innovations that defined this year’s conference. Live streaming from the authors to you.

Date, Time and Location

Nov 20, 2025 9 AM Pacific Online. Register for the Zoom!

SGBD: Sharpness-Aware Mirror Gradient with BLIP-Based Denoising for Robust Multimodal Product Recommendation

The growing integration of computer vision and machine learning into the retail industry—both online and in physical stores—has driven the adoption of multimodal recommender systems to help users navigate increasingly complex product landscapes. These systems leverage diverse data sources, such as product images, textual descriptions, and user-generated content, to better model user preferences and item characteristics. While the fusion of multimodal data helps address issues like data sparsity and cold-start problems, it also introduces challenges such as information inconsistency, noise, and increased training instability.

In this paper, we analyze these robustness issues through the lens of flat local minima and propose a strategy that incorporates BLIP—a Vision-Language Model with strong denoising capabilities—to mitigate noise in multimodal inputs. Our method, Sharpness-Aware Mirror Gradient with BLIP-Based Denoising (SGBD), is a concise yet effective training strategy that implicitly enhances robustness during optimization. Extensive theoretical and empirical evaluations demonstrate its effectiveness across various multimodal recommendation benchmarks. SGBD offers a scalable solution for improving recommendation performance in real-world retail environments, where noisy, high-dimensional, and fast-evolving product data is the norm, making it a promising paradigm for training robust multi-modal recommender systems in retail industry.

About the Speaker

Kathy Wu holds a Ph.D. in Applied Mathematics and dual M.S. degrees in Computer Science and Quantitative Finance from the University of Southern California (USC), Los Angeles, CA, USA. At USC, she served as a course lecturer, offering ML Foundations and ML for Business Applications in the science school and business school. Her academic research spans high-dimensional statistics, deep learning, and causal inference, etc.

Kathy brings industry experience from Meta, LinkedIn, and Morgan Stanley in the Bay Area and New York City, US, where she focused on AI methodologies and real-world applications. She is currently an Applied Scientist at Amazon, within the Global Store organization, leading projects in E-Commerce Recommendation Systems, Search Engines, Multi-Modal Vision-Language Models (VLMs), and LLM/GenAI in retails.

Her work has been published in top-tier conferences including ICCV, CVPR, ICLR, SIGIR, WACV, etc. At ICCV 2025, she won the Best Paper Award in Retail Vision.

Spatial Mental Modeling from Limited Views

Can VLMs imagine the unobservable space from just a few views, like humans do? Humans form spatial mental models, as internal representations of "unseen space" to reason about layout, perspective, and motion. On our proposed MINDCUBE, we see critical gap systematically on VLMs building robust spatial mental models through representing positions (cognitive mapping), orientations (perspective-taking), and dynamics (mental simulation for ''what-if'' movements). We then explore three approaches to help VLMs approximate spatial mental models, including unseen intermediate views, natural language reasoning chains, and cognitive maps.

The significant improvement comes from ''map-then-reason'' that jointly trains the model to first abstract a cognitive map and then reason upon it. By training models to construct and reason over these internal maps, we boosted accuracy from 37.8% to 60.8% (+23.0%). Adding reinforcement learning pushed performance even further to 70.7% (+32.9%). Our key insight is that such scaffolding of spatial mental models, actively constructing and utilizing internal structured spatial representations with flexible reasoning processes, significantly improves understanding of "unobservable space".

We aim to understand why geometric concepts remain challenging for VLMs and outlining promising research directions towards fostering more robust spatial intelligence.

About the Speaker

Manling Li is an Assistant Professor at Northwestern University and Amazon Scholar. She was a postdoc at Stanford University, and obtained the PhD degree in Computer Science at University of Illinois Urbana-Champaign in 2023. She works on the intersection of language, vision, and robotics, recognized by the MIT TR 35 Under 35, ACL Inaugural Dissertation Award Honorable Mention, ACL’24 Outstanding Paper Award, ACL'20 Best Demo Paper Award, and NAACL'21 Best Demo Paper Award, Microsoft Research PhD Fellowship, EE CS Rising Star, etc.

Forecasting and Visualizing Air Pollution via Sky Images and VLM-Guided Generative Models

Air pollution monitoring is traditionally limited by costly sensors and sparse data coverage. Our research introduces a vision-language model framework that predicts air quality directly from real-world sky images and also simulates skies under varying pollution levels to enhance interpretability and robustness. We further develop visualization techniques to make predictions more understandable for policymakers and the public. This talk will present our methodology, key findings, and implications for sustainable urban environments.

About the Speaker

Mohammad Saleh Vahdatpour is a PhD candidate in Computer Science at Georgia State University specializing in deep learning, vision–language models, and sustainable AI systems. His research bridges generative AI, environmental monitoring, and motion perception, focusing on scalable and energy-efficient models that connect scientific innovation with real-world impact.

Sari Sandbox: A Virtual Retail Store Environment for Embodied AI Agents

We present Sari Sandbox, a high-fidelity, photorealistic 3D retail store simulation for benchmarking embodied agents against human performance in shopping tasks. Addressing a gap in retail-specific sim environments for embodied agent training, Sari Sandbox features over 250 interactive grocery items across three store configurations, controlled via an API. It supports both virtual reality (VR) for human interaction and a vision language model (VLM)-powered embodied agent.

We also introduce SariBench, a dataset of annotated human demonstrations across varied task difficulties. Our sandbox enables embodied agents to navigate, inspect, and manipulate retail items, providing baselines against human performance. We conclude with benchmarks, performance analysis, and recommendations for enhancing realism and scalability.

About the Speakers

Emmanuel G. Maminta is a fourth-year Artificial Intelligence Ph.D. student at the Ubiquitous Computing Laboratory (UCL) in the University of the Philippines Diliman, advised by Prof. Rowel O. Atienza.

Janika Deborah B.Gajo is an undergraduate student studying for a Bachelor of Science in Computer Engineering at the University of the Philippines, Diliman.

In a rapidly evolving advertising landscape where data, technology, and methodology converge, the pursuit of rigorous yet actionable marketing measurement is more critical—and complex—than ever. This talk will showcase how modern marketers and applied data scientists employ advanced measurement approaches—such as Marketing Mix Modeling (frequentist and Bayesian) and robust experimental designs, including randomized control trials and synthetic control-based counterfactuals—to drive causal inference in advertising effectiveness for meaningful business impact.

The talk will also address emergent aspects of applied marketing science- namely open-source methodologies, digital commerce platforms and artificial intelligence usage. Innovations from industry giants like Google and Meta, as well as open-source communities exemplified by PyMC-Marketing, have democratized access to advancement in methodologies. The emergence of digital commerce platforms such as Amazon and Walmart and the rich data they bring forward is transforming how customer journeys and campaign effectiveness are measured across channels. Artificial Intelligence is accelerating every facet of the data science workflow, streamlining processes like coding, modeling, and rapid prototyping (“vibe coding”) to enabling the integration of neural networks and deep learning techniques into traditional MMM toolkits. Collectively, these provide new and easy ways of quick experimentation and learning of complex nonlinear dynamics and hidden patterns in marketing data

Bringing these threads together, the talk will show how Ovative Group—a media and marketing technology firm—integrates domain expertise, open-source solutions, strategic partnerships, and AI automation into comprehensive measurement solutions. Attendees will gain practical insights on bridging academic rigor with business relevance, empowering careers in applied data science, and helping organizations turn marketing analytics into clear, actionable strategies.

Causal inference offers a principled way to estimate the effects of interventions—a critical need in industrial settings where decisions directly impact costs and performance. This talk presents a case study from Saint-Gobain, in collaboration with Inria, where we applied causal inference methods to production and quality data to reduce raw material usage without compromising product quality. We’ll walk through each step of a causal analysis: building a causal graph in collaboration with domain experts, identifying confounders, working with continuous treatments, and using open-source tools such as DoWhy, EconML, and DAGitty. The talk is aimed at data scientists with basic ML experience, looking to apply causal thinking to real-world, non-academic problems.

This talk dives into the challenge of measuring the causal impact of app installs on customer loyalty and value, a question at the heart of data-driven marketing. While randomized controlled trials are the gold standard, they’re rarely feasible in this context. Instead, we’ll explore how observational causal inference methods can be thoughtfully applied to estimate incremental value with careful consideration of confounding, selection, and measurement biases. This session is designed for data scientists, marketing analysts, and applied researchers with a working knowledge of statistics and causal inference concepts. We’ll keep the tone practical and informative, focusing on real-world challenges and solutions rather than heavy mathematical derivations.

Attendees will learn: * How to design robust observational studies for business impact * Strategies for covariate selection and bias mitigation * The use of multiple statistical and design-based causal inference approaches * Methods for validating and refuting causal claims in the absence of true randomization We’ll share actionable insights, code snippets, and a GitHub repository with example workflows so you can apply these techniques in your own organization. By the end of the talk, you’ll be equipped to design more transparent and credible causal studies-and make better decisions about where to invest your marketing dollars.

Requirements:
A basic understanding of causal inference and Python is recommended. Materials and relevant links will be shared during the session

When and how to apply causal inference as a marketing measurement method. Examples and applications. – Irina Brudaru

​Outline:

• ​Start with a problem and brainstorm how to measure.
• Say, big posters on big buildings: how to track impact?
• Causal Inference 101 minimal theory.
• Data and Demo on the Infer (low code) platform, output and interpretation

​About the speaker:

​With a lifelong passion for coding, which began at age 12, Irina built on her engineering background and earned a master’s degree focused on big data algorithms. Over the past 15 years, she's developed into a versatile data generalist and has spent a decade mentoring aspiring professionals. A former Googler, she now serves as a data “Swiss knife” specializing in Marketing Mix Modeling (MMM) and causal inference within the sexual wellness industry. Outside of data, she’s a proud “mom” to three feathered kids, a cooking enthusiast, and an avid mushroom forager in her free time.

Join our slack: https://datatalks.club/slack.html

External registration required at nyhackr.

This month we have Vikram Mullachery returning to give a talk about chaos.

Thank you to NYU for hosting us.

Everybody attending must RSVP through the registration form at nyhackr. There is a charge for in-person and virtual tickets are free. Space is extremely limited and in-person registration closes at 3 PM the day of the talk.

About the Talk: This is a talk on chaos and its interaction with machine learning. Since most systems that we model are dynamical, it is important that we think about their attractors, and possible chaotic nature. And if so, how do we address them in machine learning systems? Contrariwise is it possible to model a chaotic system using machine learning systems?

About Vikram: I am a Sr. AI/ML engineer with deep expertise in social media algorithms, ranking and recommendation, natural language processing etc. My work in Bayesian neural networks, causal inference and reinforcement learning techniques have been practically applied at Meta and a few startups in the NYC area. Previously, I have led teams of varying sizes in distributed work setups. Currently, I am working on a proprietary machine learning system for cybersecurity.

The venue doors open at 6:30 PM America/New_York where we will continue enjoying pizza together (we encourage the virtual audience to have pizza as well). The talk, and livestream, begins at 7:00 PM America/New_York.

Remember, register at nyhackr.

Causal Inference in R is a comprehensive guide that introduces you to the methods and practices of determining causality in data through the lens of R programming. By navigating its pages and examples, you'll master the application of causal models and statistical approaches to real-world problems, enabling more informed data-driven decisions. What this Book will help me do Understand the principles and foundations of causal inference to identify causality in data. Apply methods like propensity score matching and instrumental variables using R. Leverage real-world case studies to analyze and resolve confounding factors and make better data claims. Harness statistical methods and R tools to address real-world data challenges innovatively. Develop a strategy for integrating causal models into decision-making workflows with confidence. Author(s) Subhajit Das, the author of Causal Inference in R, is an accomplished applied scientist with over a decade of experience in causal inference methodologies and data analysis. Subhajit is passionate about empowering learners by breaking down complex concepts into manageable, clear explanations. His expertise ensures that readers not only understand the theory behind causal inference but are also able to apply it effectively using R. Who is it for? This book is ideal for data analysts, statisticians, and researchers looking to deepen their understanding of causal inference techniques using R. Whether you're a practitioner aiming to enhance your data-driven decision-making skills or a student aspiring to tackle advanced causal analysis, this book provides pathbreaking insights. It's suitable for individuals at beginner to intermediate skill levels in data analysis, especially those in public policy, economics, and the social sciences who utilize R regularly.

To access this webinar, please register here: https://hubs.li/Q02FB9GC0

Topic: "Machine Learning Models To Interpretable Rules"

Speaker: Srikanth K S, Director, Data Science at Games24x7 Data Science Professional – A leader with hands-on technical expertise - Data Science, Causal inference, Explainable AI and model interpretability, Predictive modeling, Machine learning, Deep learning, Artificial Intelligence, recommender systems with a background in Applied mathematics, Statistics and Optimization. - At Walmart: Established disciplines as a data science leader, created data science pipelines, built models at scale in Retail areas such as Merchandising, Assortment, Personalization, Advertising platform, Supply-chain, Forecasting and Transportation alongside working with multiple stakeholders, cross-functional teams. Managed a team of data scientists, UI/UX developers, ML engineers and DevOps engineers.

Abstract: Some machine learning models are essentially decision rules with if-then-else constructs. Distillation of this knowledge into rulelists and rulesets provides an interpretable overview of the decision-making process. Explainability leads to clear idea about interventions, explanation to outliers and many more use-cases. We present a few hands-on use cases with 'imodels' (python package for rule based models) and 'tidyrules' (R package for ruleset manipulation and post-hoc reordering and pruning) along with utilities to convert the rulesets into SQL to bring them into production setting.

ODSC Links: • Get free access to more talks/trainings like this at Ai+ Training platform: https://hubs.li/H0Zycsf0 • ODSC blog: https://opendatascience.com/ • Facebook: https://www.facebook.com/OPENDATASCI • Twitter: https://twitter.com/_ODSC & @odsc • LinkedIn: https://www.linkedin.com/company/open-data-science • Slack Channel: https://hubs.li/Q02zdcSk0 • Code of conduct: https://odsc.com/code-of-conduct/

To access this webinar, please register here: https://hubs.li/Q02FB9GC0

Topic: "Machine Learning Models To Interpretable Rules"

Speaker: Srikanth K S, Director, Data Science at Games24x7 Data Science Professional – A leader with hands-on technical expertise - Data Science, Causal inference, Explainable AI and model interpretability, Predictive modeling, Machine learning, Deep learning, Artificial Intelligence, recommender systems with a background in Applied mathematics, Statistics and Optimization. - At Walmart: Established disciplines as a data science leader, created data science pipelines, built models at scale in Retail areas such as Merchandising, Assortment, Personalization, Advertising platform, Supply-chain, Forecasting and Transportation alongside working with multiple stakeholders, cross-functional teams. Managed a team of data scientists, UI/UX developers, ML engineers and DevOps engineers.

Abstract: Some machine learning models are essentially decision rules with if-then-else constructs. Distillation of this knowledge into rulelists and rulesets provides an interpretable overview of the decision-making process. Explainability leads to clear idea about interventions, explanation to outliers and many more use-cases. We present a few hands-on use cases with 'imodels' (python package for rule based models) and 'tidyrules' (R package for ruleset manipulation and post-hoc reordering and pruning) along with utilities to convert the rulesets into SQL to bring them into production setting.

ODSC Links: • Get free access to more talks/trainings like this at Ai+ Training platform: https://hubs.li/H0Zycsf0 • ODSC blog: https://opendatascience.com/ • Facebook: https://www.facebook.com/OPENDATASCI • Twitter: https://twitter.com/_ODSC & @odsc • LinkedIn: https://www.linkedin.com/company/open-data-science • Slack Channel: https://hubs.li/Q02zdcSk0 • Code of conduct: https://odsc.com/code-of-conduct/

To access this webinar, please register here: https://hubs.li/Q02FB9GC0

Topic: "Machine Learning Models To Interpretable Rules"

Speaker: Srikanth K S, Director, Data Science at Games24x7 Data Science Professional – A leader with hands-on technical expertise - Data Science, Causal inference, Explainable AI and model interpretability, Predictive modeling, Machine learning, Deep learning, Artificial Intelligence, recommender systems with a background in Applied mathematics, Statistics and Optimization. - At Walmart: Established disciplines as a data science leader, created data science pipelines, built models at scale in Retail areas such as Merchandising, Assortment, Personalization, Advertising platform, Supply-chain, Forecasting and Transportation alongside working with multiple stakeholders, cross-functional teams. Managed a team of data scientists, UI/UX developers, ML engineers and DevOps engineers.

Abstract: Some machine learning models are essentially decision rules with if-then-else constructs. Distillation of this knowledge into rulelists and rulesets provides an interpretable overview of the decision-making process. Explainability leads to clear idea about interventions, explanation to outliers and many more use-cases. We present a few hands-on use cases with 'imodels' (python package for rule based models) and 'tidyrules' (R package for ruleset manipulation and post-hoc reordering and pruning) along with utilities to convert the rulesets into SQL to bring them into production setting.

ODSC Links: • Get free access to more talks/trainings like this at Ai+ Training platform: https://hubs.li/H0Zycsf0 • ODSC blog: https://opendatascience.com/ • Facebook: https://www.facebook.com/OPENDATASCI • Twitter: https://twitter.com/_ODSC & @odsc • LinkedIn: https://www.linkedin.com/company/open-data-science • Slack Channel: https://hubs.li/Q02zdcSk0 • Code of conduct: https://odsc.com/code-of-conduct/

Title: Decoding Causality: How Economists Use Data to Uncover Causal Links

Brief about the talk: The presentation shall give a broad examination of how applied economists use data to uncover and confirm causal links. The focus will be put on the main tools used in the field of econometrics for causal inference, particularly emphasizing recent developments and emerging trends. The practicality of these tools will be demonstrated using cases on specific economic issues in Saudi Arabia.

Brief about the speaker: Abdullah Almansour is a Professor of Applied Economics at King Fahd University of Petroleum and Minerals (KFUPM). He is actively engaged in analyzing and debating economic issues in Saudi Arabia. Dr. Al-Mansour earned his PhD in Applied Economics from the University of Waterloo and has held visiting research positions at prestigious institutions such as Harvard's Growth Lab and the Oxford Institute for Energy Studies, among others. He has an extensive publication record in leading academic journals in economics and finance and a track record of advisory services in both public and private entities on economic matters

Hey, LondonR!

The event will be brought to you by Datacove. We are a Data Consultancy Team based in Brighton, with experience in founding many well-attended events for R and Python. New leadership team, but the same brilliant community.

We will be supported by the generous team, organised by Valerio Ficcadenti, at London Southbank University, in Elephant and Castle, London. Hosted by Abbie Brookes and Jeremy Horne from Datacove.

We are thrilled to be joined by Duncan Stoddard, Data Scientist at DS Analytics. DS Analytics help companies use data to optimise decision making. They specialise in marketing effectiveness, causal inference, customer lifetime value and all things Bayesian. Duncan will be enlightening us with a talk on 'Bayesian Marketing Mix Modelling in R'. He continues to explain that the talk will cover using Stan and R to build marketing models, whether it's worth building your own MMM or just using Robyn, and many other topics!

Next, we can't wait to be hearing from Andrie de Vries, Director of Product Strategy at Posit PBC! He will be sharing a talk on "Lessons learnt from Product Management, applied to Data Science". Expanding, "As a Data Scientist you build data products all the time. You may even have worked with a Product Manager to create analyses and dashboards for decision making. But are you applying the skills of product management in your data science role? In this talk Andrie provides an overview of Product Management (PM), and what he’s learnt over two decades of managing products, ranging from hardware (Psion PDAs) to software (Microsoft R Open, Posit Workbench) and hosted services (MRAN)."

After, we have lined up another excellent talk from Alex Glaser, Data Scientist at AIGTECH. Alex will be sharing a talk with us on 'Covid and Data Science'. Ah yes, that illness we are all still trying to forget... He goes on to tell us more, "We have all lived with COVID and lockdown over the last few years and, no doubt, had our own opinions about the rights and wrongs about the decisions taken by the Government. This talk won’t tell you the reasons why certain decisions were made (after all there is a public inquiry for that) but I will go over some of my experiences in the Infectious Disease Modelling team in the UK Health Security Agency. This talk will go over some of the projects during this time, the models used, results obtained and the papers published." We can't wait!

If our fascinating talks weren't enough to make you sign up to London R - let's tell you more!

The event will have plenty of opportunity to network amongst like-minded individuals in the R scene. The event will also boast nibbles and drinks. Plus merchandise sent directly from Posit. Additionally, for those who would love to join - we will be moving to a pub afterwards (TBC)!

We can't wait to see you all and meet some new faces. :)

Location Instructions: Arrive at the Business Building on Milcote Street. The room code is Floor One, Learning Lounge. If requiring assistance - the reception and our organisers will aid your arrival. What3Words: punk.single.enter

Join us as we explore causal accountability to improve customer engagement, using the advantages afforded by generative AI. Dr. Eric Newby, who brings his profound expertise in Applied Mathematics and a keen understanding of machine learning, will delve into an in-depth discussion on causal inference. He will shed light on how causality increases the predictability and effectiveness of machine learning models. The talk will explore the complexities of causality within data, offering invaluable insights into how to correctly interpret correlations and anomalies.

The understanding of these influences leads to better decision-making, improved algorithm performances, and eventually, more reliable and robust Machine Learning applications. Furthermore, Dr. Newby will guide you on how to implement fundamental changes in designing machine learning models keeping causality in mind. This includes overcoming challenges in identifying causal relationships and accurately modeling these in the machine learning context.

This event promises to be an engaging and enlightening experience for data scientists, machine learning enthusiasts, algorithm designers, software engineers, and all others interested in the interplay of causality in machine learning and wishing to improve the efficiency and effectiveness of their models and applications.

IAQF & Thalesians Seminar Series: Leveraging Deep Learning for Multimodal Data Analysis in Credit Risk Assessment. A Seminar by Dr. Cristián Bravo.

6:00 PM Seminar Begins 7:30 PM Reception Hybrid Event:

Fordham University McNally Amphitheater 140 West 62nd Street New York, NY 10023

Free Registration! For Virtual Attendees: Please email [email protected] for the link

Abstract: Credit risk assessment is a multifaceted process in which lenders employ various measures to evaluate the risk associated with borrowers, ranging from individual consumers to large-scale companies. To achieve a comprehensive understanding of credit risk, lenders extensively analyze a wide array of data sources, encompassing images, text, social networks, time series data, and traditional financial variables. Deep learning methodologies offer significant advantages in leveraging diverse data from multiple sources to generate accurate predictions and provide valuable insights into the complex relationships inherent in these inputs. This presentation aims to explore different strategies for handling multimodal data in both consumer and corporate lending using deep learning techniques, with a particular emphasis on transformer models. The discussion will encompass the utilization of time series data, ego networks, and textual information, in conjunction with conventional financial variables. Real-world use cases will be presented to showcase the predictive gains obtained through multimodality and demonstrate the valuable insights that can be extracted from these diverse data sources. Furthermore, the talk will address the challenges and solutions associated with deploying these models in credit risk assessment. It will shed light on the potential pitfalls that can arise when working with multimodal data and outline effective approaches to mitigate these issues. By the end of the presentation, participants will have a better understanding of the power of deep learning techniques in analyzing multimodal data in this space, enabling them to make informed decisions and enhance their lending practices.

Bio: Dr. Cristián Bravo is an Associate Professor and Canada Research Chair in Banking and Insurance Analytics at the University of Western Ontario, Canada. He also serves as the Director of the Banking Analytics Lab. His research lies at the intersection of data science, analytics, and credit risk, researching how techniques such as multimodal deep learning, causal inference, and social network analysis can be used to understand relations between consumers and financial institutions. He has over 75 academic works in high-impact journals and conferences in operational research, finance, and computer science. He serves as an editorial board member in Applied Soft Computing and the Journal of Business Analytics and is the co-author of the book “Profit Driven Business Analytics”, which has sold over 6,000 copies to date. Dr. Bravo has been quoted by The Wall Street Journal, WIRED, CTV, The Toronto Star, The Globe and Mail, and Global News. He is also a regular panelist at CBC News’ Weekend Business Panel where he discusses the latest news in Banking, Finance and Artificial Intelligence. He can be reached via LinkedIn, by Twitter @CrBravoR, or through his lab website at https://thebal.ai.