hidden states
Hidden states refer to internal representations of data in a neural model that are not directly observable but embody important information learned by the model during processing and are crucial for tasks like sequential prediction.
- Breaking the Order Barrier: Off-Policy Evaluation for Confounded POMDPs
- Chain-of-Model Learning for Language Model
- DePass: Unified Feature Attributing by Simple Decomposed Forward Pass
- Hybrid Latent Reasoning via Reinforcement Learning
- Investigating Hallucinations of Time Series Foundation Models through Signal Subspace Analysis
- Remarkable Robustness of LLMs: Stages of Inference?
- Revisiting Bi-Linear State Transitions in Recurrent Neural Networks
- Unifying Attention Heads and Task Vectors via Hidden State Geometry in In-Context Learning
- What Happens During the Loss Plateau? Understanding Abrupt Learning in Transformers
- Whose Instructions Count? Resolving Preference Bias in Instruction Fine-Tuning