Recurrent Neural Networks Learn to Store and Generate Sequences using Non-Linear Representations
Paper • 2408.10920 • Published • 1
What I don't understand
Gap in knowledge is the place I learn the most from.
-
Towards Exact Computation of Inductive Bias
Paper • 2406.15941 • PublishedNote Given a model class and a specific problem, the average number of searches required to find a good model (with low test loss) reflects the quality of that model class for that particular problem and, consequently, the validity of your assumption.
-
DrawingSpinUp: 3D Animation from Single Character Drawings
Paper • 2409.08615 • Published • 19 -
Learning to Move Like Professional Counter-Strike Players
Paper • 2408.13934 • Published • 23 -
Real-Time Video Generation with Pyramid Attention Broadcast
Paper • 2408.12588 • Published • 17 -
Diffusion Models Learn Low-Dimensional Distributions via Subspace Clustering
Paper • 2409.02426 • Published -
Neural Networks Generalize on Low Complexity Data
Paper • 2409.12446 • Published • 1 -
Target Score Matching
Paper • 2402.08667 • Published • 1 -
LLM-Augmented Symbolic Reinforcement Learning with Landmark-Based Task Decomposition
Paper • 2410.01929 • Published • 2 -
Abstract Reward Processes: Leveraging State Abstraction for Consistent Off-Policy Evaluation
Paper • 2410.02172 • Published • 1 -
Life, uh, Finds a Way: Systematic Neural Search
Paper • 2410.01349 • Published • 1 -
Learning to Build by Building Your Own Instructions
Paper • 2410.01111 • Published -
Autoregressive Large Language Models are Computationally Universal
Paper • 2410.03170 • Published • 1 -
Hyperbolic Deep Reinforcement Learning
Paper • 2210.01542 • Published • 1 -
mHC: Manifold-Constrained Hyper-Connections
Paper • 2512.24880 • Published • 321