Scaling and Foundation Models Workshop @ Mila
Mon Apr 29 - Fri May 3 Agora and Auditorium 2
3:30pm - 4:00pm Amortized Inference for Aligning Large Language Models (V Shah, M Jain, T Jiralerspong, A Ryoo) slides / video
4:00pm - 4:30pm Does providing relevant in-context examples improve reasoning in LLMs? (S Joshi and A Didolkar) slides / video
Coffee/pizza break
5:00pm - 5:30pm Parallelizable auto-regressive Inference in State Space Models (D Secrieru) slides / video
5:30pm-6:00pm Paper presentation: LLaMA Pro: Progressive LLaMA with Block Expansion (N Islah) slides / video
2:30PM - 3:00 pm Scalable and Distributed Question-Answering with Retrieval Augmented Generation (Aniket Saxena) slides / video
3:00PM - 3:30 PM Spatial VLMs for Autonomous Driving (Prince Immanuel) slides / video (start: 22:19)
3:30PM - 3:45 pm break
3:45 pm - 4:15 pm Learning planning and tool using in VLMs using RL (Sarvjeet Singh Ghotra) slides / video (start: 40:10)
4:15pm - 4:45 pm Perturbing Mixtures of Experts improves Deep RL Agent performance (Johan S) slides / video (start: 01:03:20)
4:45pm - 4:30 pm Mixture-of-Depths: Dynamically allocating compute in transformer-based language models (Nizar Islah) slides
4:30pm - 5:00pm Optimizing Communication in Federated Learning by Limiting Sharing to Scalars with Pre-Trained Models (Nicolas B, Humza W) slides / video (start: 01:18:07)
1:30pm - 2:00pm Scaling Transformer Value-based Deep RL (Anthony Gosselin, Jeremy Qin, Jinghan Sun) slides / video
2:00pm - 2:30 pm Scaling Law Analysis of Climate Foundation Models (Venkatesh Ramesh, Paloma Fernandez) slides / video
2:30pm - 3:00 pm Can μP Enable Zero-Shot Learned Optimizer Transfer? (Benjamin Thérien) slides / video
3:00pm - 3:30pm pizza break
3:30pm - 4:00 pm Aligning Language Model with User Search Intent (Yuchen Hui, Congshu Zhou, Neeraj Kumar) slides / video
4:00 pm - 4:30 pm Starcaster: Probabilistic Multivariate Time Series Forecasting (Andrew Williams, Arjun Ashok) slides / video
4:30pm - 5:00pm Visual Instruction Tuning (Le Zhang) slides / video
5:00pm - 5:30pm A Survey of State Representation Learning for Deep RL (Ayoub Echchahed, Nassim El Massaudi) slides / video
10:00 - 10:30 Evaluating and Enhancing The Adversarial Robustness of Vision Language Models (Rishika Bhagwatkar) slides / video
10:30 - 11:00 Infinite Learning Rate Schedulers for continually pretraining Foundation Models (Vaibhav Singh, Paria Mehrbod, Paul Janson) slides / video
11:00 - 11:30 coffee/snacks
11:30 - 12:00 Can LLMs model agent belief change using real-world data? (Sophie Wu ) slides / video
12:00 - 12:30 Investigating the Mechanistic Causes of Hallucinations in Language Models (Meng Cao ) slides /video
12:30 - 01:15 lunch break: pizza etc
01:15 - 01:45 Diffusion Models as Priors: Regularization Techniques (Misha Barth, Sammy Sharief ) slides / video
01:45 - 02:15 Fine-tuning LLMs for Mental Health Interventions (Tanner Ducharme ) slides / video
02:15 - 02:30 break: coffee/snacks
02:30 - 03:00 Mamba vs Transformers: Comparative Study (Maxime Petrenko) slides / video
03:00 - 03:30 Can SSMs beat Transformers? (Megh, Istabrak, Jerome) slides / video
03:30 - 03:45 coffee/snacks
03:45 - 04:15 S4 and The Chomsky Hierarchy (Xaver Morin ) slides / video
04:15 - 04:45 Neural Manifold Analysis of Classification Capacity in Continual Learning (Anirudh, Dhuruva and Neeraj ) slides / video
04:45- 05:00 coffee/snacks
05:00 - 05:30 Gitchameleon: Breaking the version barrier for code generation models (Nizar Islah) slides/ video
05:30 - 06:00 Interactive Rule Induction (Xiaoyin Chen, Junlin Wang, Xinyu Yuan, Le Zhang) slides / video
09:30 - 10:00 Brain2Vec: Searching for Multimodal Brain Dynamics Representations (William Callaghan ; Darsh Kaushik ; Juan David Vargas ; Jon Pilarte) slides / video
10:00 - 10:30 Aligning LLMs to Political Ideologies (Jean-Romain Roy) slides / video
10:30 - 11:00 break
11:00 - 11:30 Towards Tokenizer-Free Multimodality (Jonathan Siu Chi Lim; Mina Beiramy) slides / video
(Mahmood Hegazy, Karoline Lippert, Tommaso) slides / video
12:00 - 12:30 Joint Bias Mitigation and Privacy Preservation Feature Representation Framework for Vision (Maxime Gevers) slides / video
12:30 - 01:00 Evaluating Multi-Modal Alignment: Ethical Considerations in Vision Language Models (Mahsan Abdoli) slides / video
01:00 - 02:00 lunch: pizza etc
02:00 - 02:30 Transfer To Time-Series Domain (Yi Cong Li) slides / video
02:30 - 03:00 Accumulate while you Communicate: Hiding Communications in Distributed LLM Training (Adel Nabli) slides / video
03:00 - 03:30 TEARS: TExtual latent Auto-encoders for Recommender Systems (Emiliano Penaloza, Shubham Gupta) slides / video
03:30 - 04:00 Boosting Medical ML: Towards training fewer models for biomedical volume segmentation (Aloys Portafaix) slides / video
04:00 - 04:15 coffee/snacks
04:30 - 05:00 Hierarchical Modeling for Generation Hindustani Vocal Music (Nithya Shikarpur) slides / video
05:00 - 05:30 How well do Audio-Visual LLMs perform in diverse audio-visual tasks? A case study (Subhrajyoti Dasgupta) slides / video
05:30 - 06:00 Gradient dissent in language model pretraining (Andrei Mircea) slides / video
06:00 - 06:30 PixelMamba: Advancing Image Restoration with Hybrid Transformer and State-Space Architectures (Moetez Kdayem) slides / video
End-of-the-semester Party!
Fri May 3, 6:30 pm in Agora