Recording of papers I read
Related to Data Distillation / Process Supervision / Convergent Learning
- Convergent Learning: Do different neural networks learn the same representations?
- Do Wide and Deep Networks Learn the Same Things? Uncovering How Neural Network Representations Vary with Width and Depth
- Similarity of Neural Network Representations Revisited (CKA)
- Knowledge distillation: A good teacher is patient and consistent
- The Platonic Representation Hypothesis
- Let’s Verify Step by Step
- Learning ReLUs via Gradient Descent
- Object Detectors Emerge in Deep Scene CNNs
- What Knowledge Gets Distilled in Knowledge Distillation?
- Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks
- ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness
Related to Align with Human / RLHF / DAP (DPO, SLiC-HF, IPO) / Data gen:
- Training language models to follow instructions with human feedback (RLHF)
- Direct Preference Optimization: Your Language Model is Secretly a Reward Model (DPO)
- SLiC-HF: Sequence Likelihood Calibration with HumanFeedback
- A General Theoretical Paradigm to Understand Learning from Human Preferences (IPO, skip)
- Direct Language Model Alignment from Online AI Feedback (OAIF)
- Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Models (SPIN)
- Self-Rewarding Language Models
- Back to Basics: Revisiting REINFORCE Style Optimization for Learning from Human Feedback in LLMs
- Finding GPT-4’s mistakes with GPT-4
- Direct Preference Optimization with an Offset
- MindStar: Enhancing Math Reasoning in Pre-trained LLMs at Inference Time
- WARM: On the Benefits of Weight Averaged Reward Models
- WARP: On the Benefits of Weight Averaged Rewarded Policies
- Scaling Laws for Reward Model Overoptimization in Direct Alignment Algorithms
- Ever: Mitigating Hallucination in Large Language Models through Real-Time Verification and Rectification
- STaR: Bootstrapping Reasoning With Reasoning
- PROMPT2MODEL: GeneratingDeployableModelsfromNaturalLanguageInstructions
- Self-Instruct: Aligning Language Models with Self-Generated Instructions
- WizardLM: Empowering Large Language Models to Follow Complex Instructions
- Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing
- Extensive Self-Contrast Enables Feedback-Free Language Model Alignment
Related to the Interpretable ML
- Text Embeddings Reveal (Almost) As Much As Text
- Learning Concise and Descriptive Attributes for Visual Recognition
- Eliminating Information Leakage in Hard Concept Bottleneck Models with Supervised, Hierarchical Concept Learning (need extra annotation)
- ArGue: Attribute-Guided Prompt Tuning for Vision-Language Models
- Visual Transformers: Token-based Image Representation and Processing for Computer Vision
- Large Language Models are Interpretable Learners
- INViTE: INterpret and Control Vision-Language Models with Text Explanations
- Attention-based Interpretability with Concept Transformers
- Neural Prototype Trees for Interpretable Fine-grained Image Recognition
- Predicting Neural Network Accuracy from Weights
Related to LLM Agnet
- BYTESIZED32: A Corpus and Challenge Task for Generating Task-Specific World Models Expressed as Text Games
- LLM-Coordination: Evaluating and Analyzing Multi-agent Coordination Abilities in Large Language Models
Related to Framework
- Learning Transferable Visual Models From Natural Language Supervision (CLIP)
- TinyCLIP: CLIP Distillation via Affinity Mimicking and Weight Inheritance
- CoCa: Contrastive Captioners are Image-Text Foundation Models
- BLIP: Bootstrapping Language-Image Pre-training for Unified Vision-Language Understanding and Generation
- BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
- Align before Fuse: Vision and Language Representation Learning with Momentum Distillation (momentum model)
- Vision Model Pre-training on Interleaved Image-Text Data via Latent Compression Learning
Related to Weak Supervision
- Debiasing Made State-of-the-art: Revisiting the Simple Seed-based Weak Supervision for Text Classification
- Weaker Than You Think: A Critical Look at Weakly Supervised Learning
Related to General ML
- Never Train from Scratch: FAIR COMPARISON OF LONGSEQUENCE MODELS REQUIRES DATA-DRIVEN PRIORS
- Faith and Fate: Limits of Transformers on Compositionality
- Automatic Discovery of Visual Circuits
- Speculative Streaming: Fast LLM Inference without Auxiliary Models
Msic
- Topological Graph Neural Networks
- Visual Chirality
- A general mechanism for perceptual decision-making in the human brain
- Cross-Lingual Transfer for Natural Language Inference via Multilingual Prompt Translator
- Enhancing Cross-lingual Natural Language Inference by Soft Prompting with Multilingual Verbalizer
- Simul-LLM: A Framework for Exploring High-Quality Simultaneous Translation with Large Language Models
Simul-LLM
- Simul-LLM: A Framework for Exploring High-Quality Simultaneous Translation with Large Language Models
- Simultaneous Machine Translation with Large Language Models (RALCP)
- Conversational SIMULMT: Efficient Simultaneous Translation with Large Language Models (reuse KV cache)
Unknown
- Fourier Features Let Networks Learn High Frequency Functions in Low Dimensional Domains
- A Theory for Emergence of Complex Skills in Language Models
传送面板已上线，我们的行动会更加迅捷