archive
Every paper Pith has read. Search by title, abstract, or pith.
554 papers in cs.NE · page 10
-
Adaptive switching lets SNNs learn multiple Atari games at once
Scalable Multi-Task Learning through Spiking Neural Networks with Adaptive Task-Switching Policy for Intelligent Autonomous Agents
-
LLMs evolve better features for tabular models
LLM-FE: Automated Feature Engineering for Tabular Data with LLMs as Evolutionary Optimizers
-
Physics-embedded networks fix nonphysical errors in composite multiscale models
Multiscale Analysis of Woven Composites Using Hierarchical Physically Recurrent Neural Networks
-
Spike timing aligns feedback weights for local gradient descent
Spike-based alignment learning solves the weight transport problem
-
Taxonomy maps 25 years of FPGA neuromorphic architectures
A Quarter of a Century of Neuromorphic Architectures on FPGAs -- an Overview
-
Benchmark suite extends 402 job shop problems with worker flexibility
A Benchmarking Suite for Flexible Job Shop Scheduling Problems with Worker Flexibility under Uncertainty
-
-
Symbolic regression recovers gravity models and finds new mobility decay law
Distilling human mobility models with symbolic regression
-
Hebbian CNN hits 75.2% CIFAR-10 accuracy matching backprop
Advancing the Biological Plausibility and Efficacy of Hebbian Convolutional Neural Networks
-
OKAEM beats transfer methods in 12 scenarios and adapts without priors
Learning Evolution via Optimization Knowledge Adaptation
-
Binarized early-exit transformer cuts size 21x and raises accuracy
BEExformer: A Fast Inferencing Binarized Transformer with Early Exits
-
Online RNNs beat transformers at long-horizon MRI motion forecasts
Frame forecasting in cine MRI using the PCA respiratory motion model: comparing recurrent neural networks trained online and transformers
-
SincKANs yield lower errors than prior KANs on PDE tasks
Sinc Kolmogorov-Arnold network and its application for solving PDEs with singularities
-
P1-KAN outperforms other KANs on irregular functions
P1-KAN: an effective Kolmogorov-Arnold network with application to hydraulic valley optimization
-
Diffusion models run evolutionary algorithms via denoising
Diffusion Models are Evolutionary Algorithms
-
SNNs beat QNNs only below 6.4% spike rate on neuromorphic hardware
Reconsidering the energy efficiency of spiking neural networks
-
MOEAs reach O(max{k,1}n ln n) runtime on any conflict degree without tuning
Proven Advantage of Multiobjective Evolutionary Algorithms for Problems with Different Degrees of Conflict
-
Mol-genes let genetic algorithms optimize drug structures
Human-level molecular optimization driven by mol-gene evolution
-
Hop analysis links real-world problem to Leading Ones features
The hop-like problem nature -- unveiling and modelling new features of real-world problems
-
AI networks converge on one shared model of reality
The Platonic Representation Hypothesis
-
SparrowSNN cuts ECG energy by 20-100x at full accuracy
SparrowSNN: A Hardware/software Co-design for Energy Efficient ECG Classification
-
Infini-attention scales LLMs to infinite context with fixed memory
Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention
-
Spiking transformer reaches 85.65% ImageNet accuracy
QKFormer: Hierarchical Spiking Transformer using Q-K Attention
-
MemFlow adapts models to new domains with under 1% compute time
MemFlow: A Lightweight Forward Memorizing Framework for Quick Domain Adaptive Feature Mapping
-
Tensor train decomposition maximizes activation in spiking neurons
Fast gradient-free activation maximization for neurons in spiking neural networks
-
LLM evolves prompts by also evolving its own mutation rules
Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution
-
Spectral expansions learn DEs on unbounded domains without grids
A Spectral Approach for Learning Spatiotemporal Neural Differential Equations
-
SOR-SNN reorganizes one spiking network into sparse pathways for continual learning
Adaptive Reorganization of Neural Pathways for Continual Learning with Spiking Neural Networks
-
Evolutionary process during generation aligns images to preferences
Prompt Evolution for Generative AI: A Classifier-Guided Approach
-
Evolutionary search recovers scientific equations from data
Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl
-
ART clustering estimates own thresholds
A Parameter-free Adaptive Resonance Theory-based Topological Clustering Algorithm Capable of Continual Learning
-
Two rules cut energy cost of learning
Competitive plasticity to reduce the energetic costs of learning
-
Multi-spike timing backprop reaches ANN accuracy in SNNs
Timing-Based Backpropagation in Spiking Neural Networks Without Single-Spike Restrictions
-
Each neuron selects which signals to store internally
Sketch of a novel approach to a neural model
-
Dynamic estimator mix raises SNN attack success by up to 91%
Attacking the Spike: On the Transferability and Security of Spiking Neural Networks to Adversarial Examples
-
Design space for diffusion models reaches FID 1.79 on CIFAR-10
Elucidating the Design Space of Diffusion-Based Generative Models
-
Scratchpads let language models solve multi-step tasks
Show Your Work: Scratchpads for Intermediate Computation with Language Models
-
Energy decay selects stable neural paths for broad task transfer
Energy Decay Network (EDeN)
-
Random forest repairs offspring to speed multi-objective evolution
Enhanced Innovized Repair Operator for Evolutionary Multi- and Many-objective Optimization
-
C++ crossover code limits GP memory to M + 2*nthreads individuals
Multi-threaded Memory Efficient Crossover in C++ for Generational Genetic Programming
-
TUDataset supplies 120+ graph benchmarks with loaders and baselines
TUDataset: A collection of benchmark datasets for learning with graphs
-
Hopfield update rule matches transformer attention
Hopfield Networks is All You Need
-
GLU variants lift Transformer quality over ReLU and GELU
GLU Variants Improve Transformer
-
Sharing keys and values across heads speeds Transformer decoding
Fast Transformer Decoding: One Write-Head is All You Need
-
SOM hybrids beat prior methods on min-max multiple-TSP
SOM-Guided Evolutionary Search for Solving MinMax Multiple-TSP
-
Energy function trains capsules to generate images unsupervised
Training products of expert capsules with mixing by dynamic routing
-
Capsules become product of expert neurons via routing energy
Training capsules as a routing-weighted product of expert neurons
-
Learning classifier system extended to autoencoding
Autoencoding with a Learning Classifier System: Initial Results
-
Adversarial Siamese net transfers cyber attack detection across networks
Semisupervised Adversarial Neural Networks for Cyber Security Transfer Learning
-
HillVallEA19 matches top entries on CEC2013 niching suite
Benchmarking HillVallEA for the GECCO 2019 Competition on Multimodal Optimization