backpropagation
Backpropagation is a widely-used algorithm for training neural networks that computes gradients of the loss function with respect to each weight by applying the chain rule, facilitating efficient updates during training.
- $\mu$PC: Scaling Predictive Coding to 100+ Layer Networks
- A Plug-and-Play Query Synthesis Active Learning Framework for Neural PDE Solvers
- Accelerating Block Coordinate Descent for LLM Finetuning via Landscape Expansion
- EUGens: Efficient, Unified and General Dense Layers
- Efficient Parametric SVD of Koopman Operator for Stochastic Dynamical Systems
- Efficient Utility-Preserving Machine Unlearning with Implicit Gradient Surgery
- Error Broadcast and Decorrelation as a Potential Artificial and Natural Learning Mechanism
- Fast constrained sampling in pre-trained diffusion models
- Learning to Flow from Generative Pretext Tasks for Neural Architecture Encoding
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning
- Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search
- PALQO: Physics-informed model for Accelerating Large-scale Quantum Optimization
- Seeing the Wind from a Falling Leaf
- SharpZO: Hybrid Sharpness-Aware Vision Language Model Prompt Tuning via Forward-Only Passes
- Steering Information Utility in Key-Value Memory for Language Model Post-Training
- Stochastic Forward-Forward Learning through Representational Dimensionality Compression
- Transferable Black-Box One-Shot Forging of Watermarks via Image Preference Models
- Unlocking Dataset Distillation with Diffusion Models
- What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functions