Search – talk-data.com

Title & Speakers	Event
Event PyData Montreal Meetup #30 (in-person \| en personne) 2024-11-07
Advancing Deep Learning and Vision Efficiency with Mamba, VMamba, and Vim. 2024-11-07 · 23:00 moetez kdayem – MLOps Engineer @ Alex Legal Transformers are foundational in deep learning but face computational inefficiencies with long sequences. Inspired by continuous systems, Mamba, a simplified sequence model that makes State Space Models parameters dynamic and uses a hardware-aware parallel algorithm, achieving up to 5× faster inference than Transformers and linear scaling in sequence length. Mamba excels in language, audio, and genomics tasks without the need for attention mechanisms or MLP blocks. Building on Mamba, it was adapted for vision tasks where challenges like position sensitivity and global context are crucial. VMamba employs Visual State-Space (VSS) blocks and a 2D Selective Scan (SS2D) module to handle visual data efficiently, setting new benchmarks in computational efficiency and performance. Similarly, Vim (Vision Mamba) uses bidirectional Mamba blocks with position embeddings, outperforming models like DeiT without relying on self-attention, highlighting the versatility of state-space models in vision applications.
Python as an orchestrator for a RAG (retrieval-augmented generation) Architecture 2024-11-07 · 23:00 jean olivier pitre – Cloud Engineer @ DataSphere Lab @McGill University Python shines in RAG (Retrieval-Augmented Generation) systems due to its efficiency in orchestrating various processes and its extensive libraries, such as LangChain and Hugging Face Transformers. The building blocks for RAG include data extraction and preprocessing, transforming data into vectors via embedding models, and using vector databases for retrieval. Python excels in setting up data pipelines for indexing, retrieval, and generation, integrating different components, and ensuring low-latency, high-efficiency real-time processing. Real-world applications of RAG systems showcase Python's benefits and challenges in implementation, demonstrating its versatility and robustness in managing complex data flows and interactions. LLM Python RAG Vector DB

PyData Montreal Meetup #27 (in-person \| en personne) 2024-03-27 · 21:30 🔔 New PyData meetup alert 🔔 📢 The #MontrealDataCommunity is back at it again with yet another awesome meetup in store for the community. On the docket: Tiago Tex Pine, from Prodigy Education Raymond Li, from ServiceNow Research We are very grateful to ServiceNow for allowing us to host this event at their offices in the Mile-Ex 🎉. IMPORTANT : The access point is 6650 Rue Saint-Urbain #500, Montreal, Quebec H2S 3G9 Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue. 📊🤖 Don’t miss this opportunity to learn from some of the brightest minds in the Montreal data space, as well as meet fellow data practitioners & enthusiasts. We’re excited to see you there l! 👋😄 AGENDA 17h30 - Open doors 18h00 - Introduction 18h10 - Talk #1 18h50 - Break, Pizza & Networking 19h10 - Talk #2 19h50 - Final notes, break and networking 21h00 - End of event TALKS *1. How Machine Learning Can Help Game Design\,* By Tiago Tex Pine Description of the talk : The behaviour of players in a video game are high-dimensional datasets, with their own characteristics and shapes, different from any other industry. When applied with the right domain knowledge of game design and behavioural science, Machine Learning techniques can do wonders to help game makers find incredible insights to improve mechanics and the metagame of a game title. *2. BigCode: Open and Responsible development of Large Language Models for code\,* By Raymond Li Description of the talk: The BigCode project is an open scientific collaboration focused on the responsible development of large language models for code. In partnership with Software Heritage, we build The Stack v2 on top of the digital commons of their source code archive. Alongside the Software Heritage data, we carefully select other high-quality data sources. We train StarCoder2 models with 3B, 7B and 15B parameters and thoroughly evaluate them on a comprehensive set of Code LLM benchmarks. We find that our models often outperform other models of similar sizes. We make the model weights and the dataset available under permissive licenses. SPEAKER BIOS Tiago Tex Pine: I am a project leader, game designer and data scientist developing video games since 2005. I've worked with and led teams of all sizes in making 30+ games, across many different genres and platforms. Designer of game economies and free-to-play games since 2011, I'm an enthusiast of applying data science and machine learning to find deep insights into the many multi-dimensional spaces of user behaviour and player progression, and predicting player choices and churn. Today, my goal is to combine the art of game design with the science of statistics and human behaviour to create games that improve people's lives. I also run a whole bunch of tabletop RPGs and Dungeons & Dragons in my spare time. Raymond Li: Raymond is a Research Engineer in the Large-Language-Models lab in ServiceNow Research. Previously at ElementAI, and now at ServiceNow Research, his work extends across a variety of NLP topics such as summarization or text-to-SQL. In the past few years, Raymond transitioned his research focus to language models, notably for code generation within the BigCode initiative.	PyData Montreal Meetup #27 (in-person \| en personne)
PyData Montreal Meetup #25 (in-person \| en personne) 2023-11-16 · 22:30 📣 Welcome to our last Meetup of 2023 at the Computer Research Institute of Montreal! 🎉 🏢 Explore the future of AI collaboration and large language models through insightful talks by our expert speakers, Clodéric Mars and Alex Kim. IMPORTANT : * The access point is on 405 Avenue Ogilvy, bureau #101, Montréal (Parc Metro - Free parking) Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue 📊🤖 Don't miss this opportunity to explore cutting-edge applications and ignite your curiosity. Save the date and secure your spot now! 👉🗓️ See you there! 👋😄 Agenda* 5:30 pm - Doors open 6:00 pm - Introduction 6:10 pm - “Human/AI superteams: building collaborative systems with human-in-the-loop learning” by Clodéric Mars 6:50 pm - Break, pizza and networking 7:10 pm - "A Gentle Introduction to Large Language Models" by Alex Kim 7:50 pm - Final notes, break and networking Talks Human/AI superteams: building collaborative systems with human-in-the-loop learning Abstract: The active involvement of humans in AI training and operation, known as « Human-in-the-loop learning" (HILL), is where the next frontier of effective and reliable decision-making systems lies. HILL techniques enable a holistic view of the training and deployment of hybrid human-IA teams in which the AI component learns from the expertise of its human teammates and continually adapts to their expectations, and where, similarly, humans can better understand the strengths and weaknesses of the AI. In this presentation, we will give an overview of some HILL techniques and their characteristics, with a closer look at reinforcement learning with human feedback (RLHF), which has recently given rise to a new generation of conversational assistants better aligned with the expectations of human users, with ChatGPT for example being the first incarnation. We will then illustrate the impact of these techniques through industrial use cases developed by AI Redefined, using Cogment, our artificial learning development solution (MLOps) for HILL methods. Finally, we'll draw up a roadmap of the challenges to be addressed if these methods are to have a wider impact. About Clodéric: Clodéric has been building AIs since 2006 with one overarching goal: fostering the collaboration between Humans and AIs. At INRIA then at Golaem and MASA Group he has worked on explicit AI techniques applied to video games, simulation and special effects. In 2015, he co-founded craft ai to focus on explainable machine learning. Since 2020, he has been working at AI Redefined on Cogment to make it accessible to create collaborative systems including humans and AIs. Over the years, he contributed to the development of AIs collaborating with artists to populate movie shots, designers to understand transportation infrastructure usage, instructors to create realistic and purposeful military training or engineers to operate energy infrastructures. "A Gentle Introduction to Large Language Models" Abstract: This presentation is tailored for both non-ML professionals curious about integrating LLMs into their company products and technical individuals new to LLMs. We'll begin with a simple overview of LLMs, touching on their history and basic definitions.We'll demystify how these models are trained, making it relatable without heavy jargon. Though LLMs are powerful, they have challenges. We'll address common hurdles, ensuring everyone understands both their potential and limits.For those wondering how LLMs can improve their productivity, we'll provide straightforward tips. We'll also discuss building LLM-powered applications, focusing on aspects like user interfaces and deciding between external or in-house solutions.Concluding, we'll peek into the future of LLMs. Join us for a clear and newby-friendly dive into the world of Large Language Models, suitable for both business and technical newcomers. About Alex: Alex is a physicist by degree and an ML engineer by trade. He's contributed to open-source projects widely used in the industry and academic studies. He has experience in developing ML-powered projects for a diverse range of companies, from emerging startups to established Fortune 500 firms. Now an independent consultant, Alex assists businesses in enhancing their ML and MLOps methodologies. In addition, he co-founded the PyData Montreal and Montreal MLOps meetup groups. Beyond this, he serves as an instructor for O'Reilly Media, where he offers courses on MLOps best practices.	PyData Montreal Meetup #25 (in-person \| en personne)
PyData Montreal Meetup #24 (in-person \| en personne) 2023-08-31 · 21:30 📣 Join us for an exciting "Back to School" meetup at the Ubisoft Montreal office! 🎉🏢 Discover the power of Python in Data Journalism and Reinforcement Learning through insightful talks by our expert speakers, Roberto Rocha and Yann Bouteiller. IMPORTANT : * The access point is on Saint Dominique side (5480 rue saint Dominique) and NOT on Boulevard Saint laurent * Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue 📊🤖 Don't miss this opportunity to explore cutting-edge applications and ignite your curiosity. Save the date and secure your spot now! 👉🗓️ See you there! 👋😄 Agenda 5:30 pm - Doors open 6:00 pm - Introduction 6:10 pm - How we used Python to see how many lobbyists attended Liberal fundraisers by Roberto Rocha (Investigative Journalism Foundation) 6:50 pm - Break, pizza and networking 7:10 pm - Using Video Games for Autonomous driving by Yann Bouteiller (Polytechnique Montréal) 7:50 pm - Final notes, break and networking Talks How we used Python to see how many lobbyists attended Liberal fundraisers Abstract: There's nothing illegal about lobbyists supporting political parties, but it's certainly suspicious when lobbyists give money to politicians they are trying to influence. We set out to understand just how many active lobbyists attended Liberal fundraising events, which can cost up to $1,700 to attend. To do this, we crossed the guest lists of fundraisers released by the Liberal Party and Elections Canada and the federal lobby registry, which we scrape daily. This involved a complicated mix of PDF parsing, fuzzy matching, pandas wrangling, and good old-fashioned gumshoe sleuthing. About Roberto: He is an investigative data journalist who applies data science and storytelling methods to uncover news that speaks truth to power. He currently works for the nonprofit startup Investigative Journalism Foundation, and before that worked for the CBC News investigative unit and the Montreal Gazette. When he's not geeking out with new data-focused Python packages, he likes playing Latin jazz guitar, making pizza, shaking cocktails, and doing AcroYoga in the park. Using Video Games for Autonomous driving Abstract: What if a recreational video game transformed into a valuable simulator for the autonomous car industry? As companies like Tesla opt out of traditional sensors and increasingly rely on vision-based Machine Learning, this idea becomes unexpectedly relevant. Being highly optimized for visual rendering, recent editions of popular games like Forza and Need For Speed offer unparalleled trade-offs between computational complexity and visual fidelity. In this presentation, we will explore the process of training a vision-based, real-time agent to drive in TrackMania 2020 (Ubisoft Nadeo). Despite being a visually less serious game, TrackMania's low-resource, free-to-play and creative nature makes it a well-suited proxy for this task. The final part of the talk will then open the discussion about the potential future of AI-driven bots and their capabilities. About Yann: He is a machine learning engineer specializing in Deep Reinforcement Learning (RL) for Robotics. Currently holding the position of Research Associate at Polytechnique Montreal, Yann has made significant contributions to the field through a diverse array of publications and open-source projects, spanning from the theoretical underpinnings of Deep RL to advancements in cognitive neuroscience. He leads the development of TMRL, an open-source framework tailored to facilitate the secure and optimized implementation of ad-hoc Deep RL pipelines for complex real-time applications. This includes autonomous driving, high-frequency trading, and other high-frequency control scenarios. Notably, TMRL garners recognition for its illustrative pipeline in the TrackMania videogame.	PyData Montreal Meetup #24 (in-person \| en personne)

Title & Speakers	Event
Event PyData Montreal Meetup #30 (in-person \| en personne) 2024-11-07
Advancing Deep Learning and Vision Efficiency with Mamba, VMamba, and Vim. 2024-11-07 · 23:00 moetez kdayem – MLOps Engineer @ Alex Legal Transformers are foundational in deep learning but face computational inefficiencies with long sequences. Inspired by continuous systems, Mamba, a simplified sequence model that makes State Space Models parameters dynamic and uses a hardware-aware parallel algorithm, achieving up to 5× faster inference than Transformers and linear scaling in sequence length. Mamba excels in language, audio, and genomics tasks without the need for attention mechanisms or MLP blocks. Building on Mamba, it was adapted for vision tasks where challenges like position sensitivity and global context are crucial. VMamba employs Visual State-Space (VSS) blocks and a 2D Selective Scan (SS2D) module to handle visual data efficiently, setting new benchmarks in computational efficiency and performance. Similarly, Vim (Vision Mamba) uses bidirectional Mamba blocks with position embeddings, outperforming models like DeiT without relying on self-attention, highlighting the versatility of state-space models in vision applications.
Python as an orchestrator for a RAG (retrieval-augmented generation) Architecture 2024-11-07 · 23:00 jean olivier pitre – Cloud Engineer @ DataSphere Lab @McGill University Python shines in RAG (Retrieval-Augmented Generation) systems due to its efficiency in orchestrating various processes and its extensive libraries, such as LangChain and Hugging Face Transformers. The building blocks for RAG include data extraction and preprocessing, transforming data into vectors via embedding models, and using vector databases for retrieval. Python excels in setting up data pipelines for indexing, retrieval, and generation, integrating different components, and ensuring low-latency, high-efficiency real-time processing. Real-world applications of RAG systems showcase Python's benefits and challenges in implementation, demonstrating its versatility and robustness in managing complex data flows and interactions. LLM Python RAG Vector DB

PyData Montreal Meetup #27 (in-person \| en personne) 2024-03-27 · 21:30 🔔 New PyData meetup alert 🔔 📢 The #MontrealDataCommunity is back at it again with yet another awesome meetup in store for the community. On the docket: Tiago Tex Pine, from Prodigy Education Raymond Li, from ServiceNow Research We are very grateful to ServiceNow for allowing us to host this event at their offices in the Mile-Ex 🎉. IMPORTANT : The access point is 6650 Rue Saint-Urbain #500, Montreal, Quebec H2S 3G9 Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue. 📊🤖 Don’t miss this opportunity to learn from some of the brightest minds in the Montreal data space, as well as meet fellow data practitioners & enthusiasts. We’re excited to see you there l! 👋😄 AGENDA 17h30 - Open doors 18h00 - Introduction 18h10 - Talk #1 18h50 - Break, Pizza & Networking 19h10 - Talk #2 19h50 - Final notes, break and networking 21h00 - End of event TALKS *1. How Machine Learning Can Help Game Design\,* By Tiago Tex Pine Description of the talk : The behaviour of players in a video game are high-dimensional datasets, with their own characteristics and shapes, different from any other industry. When applied with the right domain knowledge of game design and behavioural science, Machine Learning techniques can do wonders to help game makers find incredible insights to improve mechanics and the metagame of a game title. *2. BigCode: Open and Responsible development of Large Language Models for code\,* By Raymond Li Description of the talk: The BigCode project is an open scientific collaboration focused on the responsible development of large language models for code. In partnership with Software Heritage, we build The Stack v2 on top of the digital commons of their source code archive. Alongside the Software Heritage data, we carefully select other high-quality data sources. We train StarCoder2 models with 3B, 7B and 15B parameters and thoroughly evaluate them on a comprehensive set of Code LLM benchmarks. We find that our models often outperform other models of similar sizes. We make the model weights and the dataset available under permissive licenses. SPEAKER BIOS Tiago Tex Pine: I am a project leader, game designer and data scientist developing video games since 2005. I've worked with and led teams of all sizes in making 30+ games, across many different genres and platforms. Designer of game economies and free-to-play games since 2011, I'm an enthusiast of applying data science and machine learning to find deep insights into the many multi-dimensional spaces of user behaviour and player progression, and predicting player choices and churn. Today, my goal is to combine the art of game design with the science of statistics and human behaviour to create games that improve people's lives. I also run a whole bunch of tabletop RPGs and Dungeons & Dragons in my spare time. Raymond Li: Raymond is a Research Engineer in the Large-Language-Models lab in ServiceNow Research. Previously at ElementAI, and now at ServiceNow Research, his work extends across a variety of NLP topics such as summarization or text-to-SQL. In the past few years, Raymond transitioned his research focus to language models, notably for code generation within the BigCode initiative.	PyData Montreal Meetup #27 (in-person \| en personne)
PyData Montreal Meetup #25 (in-person \| en personne) 2023-11-16 · 22:30 📣 Welcome to our last Meetup of 2023 at the Computer Research Institute of Montreal! 🎉 🏢 Explore the future of AI collaboration and large language models through insightful talks by our expert speakers, Clodéric Mars and Alex Kim. IMPORTANT : * The access point is on 405 Avenue Ogilvy, bureau #101, Montréal (Parc Metro - Free parking) Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue 📊🤖 Don't miss this opportunity to explore cutting-edge applications and ignite your curiosity. Save the date and secure your spot now! 👉🗓️ See you there! 👋😄 Agenda* 5:30 pm - Doors open 6:00 pm - Introduction 6:10 pm - “Human/AI superteams: building collaborative systems with human-in-the-loop learning” by Clodéric Mars 6:50 pm - Break, pizza and networking 7:10 pm - "A Gentle Introduction to Large Language Models" by Alex Kim 7:50 pm - Final notes, break and networking Talks Human/AI superteams: building collaborative systems with human-in-the-loop learning Abstract: The active involvement of humans in AI training and operation, known as « Human-in-the-loop learning" (HILL), is where the next frontier of effective and reliable decision-making systems lies. HILL techniques enable a holistic view of the training and deployment of hybrid human-IA teams in which the AI component learns from the expertise of its human teammates and continually adapts to their expectations, and where, similarly, humans can better understand the strengths and weaknesses of the AI. In this presentation, we will give an overview of some HILL techniques and their characteristics, with a closer look at reinforcement learning with human feedback (RLHF), which has recently given rise to a new generation of conversational assistants better aligned with the expectations of human users, with ChatGPT for example being the first incarnation. We will then illustrate the impact of these techniques through industrial use cases developed by AI Redefined, using Cogment, our artificial learning development solution (MLOps) for HILL methods. Finally, we'll draw up a roadmap of the challenges to be addressed if these methods are to have a wider impact. About Clodéric: Clodéric has been building AIs since 2006 with one overarching goal: fostering the collaboration between Humans and AIs. At INRIA then at Golaem and MASA Group he has worked on explicit AI techniques applied to video games, simulation and special effects. In 2015, he co-founded craft ai to focus on explainable machine learning. Since 2020, he has been working at AI Redefined on Cogment to make it accessible to create collaborative systems including humans and AIs. Over the years, he contributed to the development of AIs collaborating with artists to populate movie shots, designers to understand transportation infrastructure usage, instructors to create realistic and purposeful military training or engineers to operate energy infrastructures. "A Gentle Introduction to Large Language Models" Abstract: This presentation is tailored for both non-ML professionals curious about integrating LLMs into their company products and technical individuals new to LLMs. We'll begin with a simple overview of LLMs, touching on their history and basic definitions.We'll demystify how these models are trained, making it relatable without heavy jargon. Though LLMs are powerful, they have challenges. We'll address common hurdles, ensuring everyone understands both their potential and limits.For those wondering how LLMs can improve their productivity, we'll provide straightforward tips. We'll also discuss building LLM-powered applications, focusing on aspects like user interfaces and deciding between external or in-house solutions.Concluding, we'll peek into the future of LLMs. Join us for a clear and newby-friendly dive into the world of Large Language Models, suitable for both business and technical newcomers. About Alex: Alex is a physicist by degree and an ML engineer by trade. He's contributed to open-source projects widely used in the industry and academic studies. He has experience in developing ML-powered projects for a diverse range of companies, from emerging startups to established Fortune 500 firms. Now an independent consultant, Alex assists businesses in enhancing their ML and MLOps methodologies. In addition, he co-founded the PyData Montreal and Montreal MLOps meetup groups. Beyond this, he serves as an instructor for O'Reilly Media, where he offers courses on MLOps best practices.	PyData Montreal Meetup #25 (in-person \| en personne)
PyData Montreal Meetup #24 (in-person \| en personne) 2023-08-31 · 21:30 📣 Join us for an exciting "Back to School" meetup at the Ubisoft Montreal office! 🎉🏢 Discover the power of Python in Data Journalism and Reinforcement Learning through insightful talks by our expert speakers, Roberto Rocha and Yann Bouteiller. IMPORTANT : * The access point is on Saint Dominique side (5480 rue saint Dominique) and NOT on Boulevard Saint laurent * Get your proof of attendance with you (available on the meetup event page or received per email), for us to check your status (waiting list or not) at the venue 📊🤖 Don't miss this opportunity to explore cutting-edge applications and ignite your curiosity. Save the date and secure your spot now! 👉🗓️ See you there! 👋😄 Agenda 5:30 pm - Doors open 6:00 pm - Introduction 6:10 pm - How we used Python to see how many lobbyists attended Liberal fundraisers by Roberto Rocha (Investigative Journalism Foundation) 6:50 pm - Break, pizza and networking 7:10 pm - Using Video Games for Autonomous driving by Yann Bouteiller (Polytechnique Montréal) 7:50 pm - Final notes, break and networking Talks How we used Python to see how many lobbyists attended Liberal fundraisers Abstract: There's nothing illegal about lobbyists supporting political parties, but it's certainly suspicious when lobbyists give money to politicians they are trying to influence. We set out to understand just how many active lobbyists attended Liberal fundraising events, which can cost up to $1,700 to attend. To do this, we crossed the guest lists of fundraisers released by the Liberal Party and Elections Canada and the federal lobby registry, which we scrape daily. This involved a complicated mix of PDF parsing, fuzzy matching, pandas wrangling, and good old-fashioned gumshoe sleuthing. About Roberto: He is an investigative data journalist who applies data science and storytelling methods to uncover news that speaks truth to power. He currently works for the nonprofit startup Investigative Journalism Foundation, and before that worked for the CBC News investigative unit and the Montreal Gazette. When he's not geeking out with new data-focused Python packages, he likes playing Latin jazz guitar, making pizza, shaking cocktails, and doing AcroYoga in the park. Using Video Games for Autonomous driving Abstract: What if a recreational video game transformed into a valuable simulator for the autonomous car industry? As companies like Tesla opt out of traditional sensors and increasingly rely on vision-based Machine Learning, this idea becomes unexpectedly relevant. Being highly optimized for visual rendering, recent editions of popular games like Forza and Need For Speed offer unparalleled trade-offs between computational complexity and visual fidelity. In this presentation, we will explore the process of training a vision-based, real-time agent to drive in TrackMania 2020 (Ubisoft Nadeo). Despite being a visually less serious game, TrackMania's low-resource, free-to-play and creative nature makes it a well-suited proxy for this task. The final part of the talk will then open the discussion about the potential future of AI-driven bots and their capabilities. About Yann: He is a machine learning engineer specializing in Deep Reinforcement Learning (RL) for Robotics. Currently holding the position of Research Associate at Polytechnique Montreal, Yann has made significant contributions to the field through a diverse array of publications and open-source projects, spanning from the theoretical underpinnings of Deep RL to advancements in cognitive neuroscience. He leads the development of TMRL, an open-source framework tailored to facilitate the secure and optimized implementation of ad-hoc Deep RL pipelines for complex real-time applications. This includes autonomous driving, high-frequency trading, and other high-frequency control scenarios. Notably, TMRL garners recognition for its illustrative pipeline in the TrackMania videogame.	PyData Montreal Meetup #24 (in-person \| en personne)

Activities & events