archive
Every paper Pith has read. Search by title, abstract, or pith.
554 papers in cs.NE · page 3
-
Future behaviors define functional structure without semantics
Canonical Functionalism: Defining Functional Structure without Observer-Relative Semantic Maps
-
Vortex-based optimizer tops 34 of 58 CEC 2017 cases
Drain-Vortex Optimization: A Population-Based Metaheuristic Inspired by Multi-Drain Free-Vortex Flow
-
4B agent matches larger models at heuristic design
AHD Agent: Agentic Reinforcement Learning for Automatic Heuristic Design
-
Finite point evaluations reconstruct convex Lipschitz functionals
Structure-Preserving Reconstruction of Convex Lipschitz Functionals on Hilbert Spaces from Finite Samples
-
Parameter reconstruction trains spiking networks to global optimality
Globally Optimal Training of Spiking Neural Networks via Parameter Reconstruction
-
Duffing ring breaks symmetry to read input shape
Broken-symmetry shape discrimination on a driven Duffing ring
-
LLM multi-agent system recovers ODEs more accurately
Discovering Ordinary Differential Equations with LLM-Based Qualitative and Quantitative Evaluation
-
Preprocessing changes flip EEG predictions in up to 42% of trials
Same Brain, Different Prediction: How Preprocessing Choices Undermine EEG Decoding Reliability
-
Direct-coded SNNs convert to event-based for lower energy use
Direct-to-Event Spiking Neural Network Transfer
-
Fixed neural networks with definable layers have finite PAC sample complexity
Every Feedforward Neural Network Definable in an o-Minimal Structure Has Finite Sample Complexity
-
Genetic search keeps research agents improving past local optima
GEAR: Genetic AutoResearch for Agentic Code Evolution
-
Diffusion, score, and flow models share one vector field
A Unified Measure-Theoretic View of Diffusion, Score-Based, and Flow Matching Generative Models
-
Causal emergence in RL agents predicts their final reward
The Causally Emergent Alignment Hypothesis: Causal Emergence Aligns with and Predicts Final Reward in Reinforcement Learning Agents
-
Decomposition strategies boost LLM heuristics for coupled optimization
CoupleEvo: Evolving Heuristics for Coupled Optimization Problems Using Large Language Models
-
Asynchronous updates match synchronous recall in kernel Hopfield networks
Efficient event-driven retrieval in high-capacity kernel Hopfield networks
-
Tuned kernels let async Hopfield nets match sync retrieval
Efficient event-driven retrieval in high-capacity kernel Hopfield networks
-
Momentum reordering enables parallel delta linear attention
MDN: Parallelizing Stepwise Momentum for Delta Linear Attention
-
Graph Normalization converges to binary MWIS solutions
Graph Normalization: Fast Binarizing Dynamics for Differentiable MWIS
-
S-LCG reaches within 1% of optimum in 83% of benchmark cases
S-LCG: Structured Linear Congruential Generator-Based Deterministic Algorithm for Search and Optimization
-
Optimizers simulate Darwinian evolution with added drift noise
Direct From Darwin: Deriving Advanced Optimizers From Evolutionary First Principles
-
Optimizers like SGD derive from Darwinian evolution with added noise
Direct From Darwin: Deriving Advanced Optimizers From Evolutionary First Principles
-
PIAS beats single best even with 25% budget on features
On the Influence of the Feature Computation Budget on Per-Instance Algorithm Selection for Black-Box Optimization
-
Lightweight 3D U-Net hits 0.727 Dice on brain tumors with 2.22M params
DALight-3D: A Lightweight 3D U-Net for Brain Tumor Segmentation from Multi-Modal MRI
-
V1 decodes images using coarse structure over details
Interpreting V1 Population Activity via Image-Neural Latent Representation Alignment
-
Hormone loops let small AI models reason deeply
S-AI-Recursive: A Bio-Inspired and Temporal Sparse AI Architecture for Iterative, Introspective, and Energy-Frugal Reasoning
-
Adaptive queries cut decision regret 25% on hard problems
QUIVER: Cost-Aware Adaptive Preference Querying in Surrogate-Assisted Evolutionary Multi-Objective Optimization
-
Control plane unifies physical neural networks across materials
phys-MCP: A Control Plane for Heterogeneous Physical Neural Networks
-
Exact algorithm computes full Pareto front for grid topology in minutes
Exact and Evolutionary Algorithms for Sequential Multi-Objective Transmission Topology Planning
-
Multi-hop graph paths recover RNN timing and guide better regularization
Unifying Dynamical Systems and Graph Theory to Mechanistically Understand Computation in Neural Networks
-
Resolvent regularization aligns RNN sparsity to task structure
Unifying Dynamical Systems and Graph Theory to Mechanistically Understand Computation in Neural Networks
-
Wedding matchmaking model guides faster optimization search
Indian Wedding System Optimization (IWSO): A Novel Socially Inspired Metaheuristic with Operational Design and Analysis
-
Physics ODE layers match Neural ODEs on housing data with fewer parameters
Physics-Modeled Neural Networks
-
Exact symmetry forces dim(G/H) zero-growth directions in recurrent dynamics
Symmetry-Protected Lyapunov Neutral Modes in Equivariant Recurrent Networks
-
Genetic algorithm evolves neural net weights for Minecraft navigation
Neuromorphic Control for 3D Navigation in Minecraft Using Genetic Algorithms
-
One spiking transformer resizes itself for different hardware budgets
Elastic Spiking Transformers for Efficient Gesture Understanding
-
MPCS reaches 94.2 efficiency score on 31-task continual learning benchmark
MPCS: Neuroplastic Continual Learning via Multi-Component Plasticity and Topology-Aware EWC
-
Review finds RL model reuse succeeds with task similarity or alignment
Combining Trained Models in Reinforcement Learning
-
The paper surveys neural architecture search methods through the lens of efficiency
HERCULES: Hardware-Efficient, Robust, Continual Learning Neural Architecture Search
-
SNN filter cuts DVS noise hardware to 11% memory
SNNF: An SNN-based Near-Sensor Noise Filter for Dynamic Vision Sensors
-
PolyStep trains spiking networks to 93.4% accuracy without gradients
Training Non-Differentiable Networks via Optimal Transport
-
ShiftLIF matches multi-level accuracy at near-binary energy cost
ShiftLIF: Efficient Multi-Level Spiking Neurons with Power-of-Two Quantization
-
Rank-one alignment erases memorization below chance
Probe-Geometry Alignment: Erasing the Cross-Sequence Memorization Signature Below Chance
-
Bayesian-Hebbian rules top associative memory capacity benchmarks
Benchmarking local Hebbian learning rules for memory storage and prototype extraction
-
Volatility updates in HGF now avoid negative precision
Robust volatility updates for Hierarchical Gaussian Filtering
-
LLMs steer distributed agents to better consensus with less talk
Learning to Act and Cooperate for Distributed Black-Box Consensus Optimization
-
Spiking machines and transformers share five sequence operations
Spiking Sequence Machines and Transformers
-
Three gate tweaks fix MoE routing at domain transitions
Affinity Is Not Enough: Recovering the Free Energy Principle in Mixture-of-Experts
-
Fixed feedback enables local learning in recurrent SNNs
Scalable Learning in Structured Recurrent Spiking Neural Networks without Backpropagation
-
Noise stability, not geometry, caps kernel Hopfield storage
Geometric and dynamical analysis of attractor boundaries and storage limits in kernel Hopfield networks
-
Kernel Hopfield storage limits arise from dynamical instability
Geometric and dynamical analysis of attractor boundaries and storage limits in kernel Hopfield networks