• Is Pre-training Truly Better Than Meta-Learning?
• Recurrent Vision Transformers for Object Detection with Event Cameras
• Supervised Pretraining Can Learn In-Context Reinforcement Learning
• Faster Segment Anything: Towards Lightweight SAM for Mobile Applications
• Kosmos-2: Grounding Multimodal Large Language Models to the World
• Aligning Large Multi-Modal Model with Robust Instruction Tuning

Support the show

• Bring Your Own Data! Self-Supervised Evaluation for Large Language Models
• Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold
• System-Level Natural Language Feedback
• Scaling MLPs: A Tale of Inductive Bias

Support the show

• ClinicalGPT: Large Language Models Finetuned with Diverse Medical Data and Comprehensive Evaluation
• LMFlow: An Extensible Toolkit for Finetuning and Inference of Large Foundation Models
• Tensor Programs V: Tuning Large Neural Networks via Zero-Shot Hyperparameter Transfer

Support the show

• Fine-Tuning Language Models with Just Forward Passes
• Opportunities and Risks of LLMs for Scalable Deliberation with Polis
• Adam Accumulation to Reduce Memory Footprints of both Activations and Gradients for Large-scale DNN Training
• Multitrack Music Transcription with a Time-Frequency Perceiver

Support the show

• AudioPaLM: A Large Language Model That Can Speak and Listen
• SequenceMatch: Imitation Learning for Autoregressive Sequence Modelling with Backtracking
• Any Deep ReLU Network is Shallow
• Constant Memory Attention Block
• From Word Models to World Models: Translating from Natural Language to the Probabilistic Language of Thought
• Quantizable Transformers: Removing Outliers by Helping Attention Heads Do Nothing

Support the show

• Fast Segment Anything
• Learning to Generate Better Than Your LLM
• DynaQuant: Compressing Deep Learning Training Checkpoints via Dynamic Quantization
• Training Transformers with 4-bit Integers

Support the show

• Textbooks Are All You Need
• vLLM: Easy, Fast, and Cheap LLM Serving with PagedAttention
• IndicTrans2: Towards High-Quality and Accessible Machine Translation Models for all 22 Scheduled Indian Languages
• Beyond Black Box AI-Generated Plagiarism Detection: From Sentence to Document Level

Support the show

• Demystifying GPT Self-Repair for Code Generation
• RoboCat: A Self-Improving Foundation Agent for Robotic Manipulation
• Wuerstchen: Efficient Pretraining of Text-to-Image Models
• Faith and Fate: Limits of Transformers on Compositionality

Support the show

Papers:

• Inverse Scaling: When Bigger Isn't Better
• Gradient is All You Need?
• Full Parameter Fine-tuning for Large Language Models with Limited Resources
• Scaling Data-Constrained Language Models
• AQuA: A Benchmarking Tool for Label Quality Assessment

Support the show

• Are ChatGPT and Other Similar Systems the Modern Lernaean Hydras of AI?
• Infinite Photorealistic Worlds using Procedural Generation
• SqueezeLLM: Dense-and-Sparse Quantization
• Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture

Support the show