ITQ3_S: 3-bit weight quantization method for LLM inference using rotation-domain adaptive quantization to reduce precision loss from weight distribution outliers.
Proteina-Complexa: fully atomistic protein binder generation method combining conditional generative modeling with structure-based optimization.
Optimization techniques for efficient inference in large vision-language models addressing computational bottlenecks from high-resolution visual tokens.
Distributed stochastic gradient descent with game-theoretic incentives to prevent gradient manipulation by strategic agents while ensuring convergence.
Principal Prototype Analysis on Manifold: interpretability method for reinforcement learning agents using prototype-based explanations.
Generalization of diffusion models to correlated stochastic sampling using probabilistic computers beyond standard neural network implementations.
FedDES: graph-based dynamic ensemble selection for personalized federated learning that addresses negative transfer through selective peer integration.
Analysis of diffusion maps showing they provide spectral representation of geometry rather than dimensionality reduction, compared with Isomap and UMAP.
ROVED: hybrid reinforcement learning framework combining vision-language embeddings with oracle feedback to reduce annotation costs for reward learning.
Koopman-based surrogate models for RL control of fluid dynamics with mitigation of distribution shifts.
InkDrop demonstrates backdoor attacks against dataset condensation methods through invisible trigger implantation.
Heddle is a distributed system for orchestrating agentic RL rollouts with LLMs to address trajectory generation bottlenecks.
Training methodology to verify and bound Lipschitz constants of neural networks for adversarial robustness and generalization.
GVF framework models health risk as vector fields on simplicial complexes from multimodal wearable and environmental data.
Federated learning approach for livestock growth prediction addressing privacy concerns and limited data availability.
ORACAL is a multimodal GNN framework for smart contract vulnerability detection with causal graphs and explainability.
Global-regional coupling framework using Transformers for kilometer-scale regional weather forecasting.
PCGS framework for strictly online prediction under non-stationarity with Transformer instantiation for expert switching.
Perturbation-based approach for unconstrained bandit linear optimization with improved regret guarantees.
ERPO uses token-level entropy-regulated policy optimization to improve credit assignment in reinforcement learning for language models.
Variational neurons in Transformer feed-forward layers to incorporate uncertainty into internal computation for language modeling.
MR-CDM framework for multi-resolution time series generation using hierarchical decomposition and diffusion models.
Study on robustness against data corruption in offline multi-agent reinforcement learning from human feedback.
Framework for estimating learning complexity and communication costs in federated learning systems before deployment.
FI-KAN introduces fractal interpolation function bases into Kolmogorov-Arnold Networks for improved multi-scale function approximation.
arXiv paper proposing optical in-network computing to reduce communication overhead in distributed machine learning systems.
arXiv paper introducing LIBERO-Para benchmark to evaluate robustness of Vision-Language-Action models to paraphrased instructions in robotic tasks.
arXiv paper proposing FedRCO, a second-order optimization framework for federated learning with improved stability under non-IID data.
arXiv paper addressing fairness issues in graph condensation, preventing amplification of demographic biases during dataset compression.
arXiv paper integrating learning-based optimization with classical statistical methods for efficient high-dimensional matrix estimation.
arXiv paper applying deep reinforcement learning to maritime coverage path planning on irregular hexagonal grids.
arXiv paper addressing label-efficient retraining of malware detection models under distribution drift in real-world settings.
arXiv paper on Bayesian framework for preference learning in many-objective optimization using mixture models of latent preference archetypes.
arXiv paper presenting evolutionary framework using LLMs to discover novel reinforcement learning algorithms by searching over executable update rules.
arXiv paper introducing KGroups, a feature selection algorithm for high-dimensional biological data using max-relevance min-redundancy criteria.
IsoQuant uses quaternion algebra and isoclinic rotations for efficient LLM KV cache compression with hardware-aligned blockwise operations.
FeDMRA addresses federated class-incremental learning with dynamic memory replay for non-IID distributed healthcare data.
HISA improves efficiency of token-level sparse attention mechanisms through hierarchical indexing, reducing O(L²) bottleneck.
Analysis of scaling laws in AI across model families, explaining their predictive power and universal effectiveness in training loss reduction.
CirrusBench evaluates LLM-based agents in real-world cloud service environments beyond correctness, measuring robustness and efficiency.
Simplex denoising framework for discrete generative modeling using non-Markovian noising scheme, applied to graph generation.
Offline multi-agent reinforcement learning approach using Partial Action Replacement to handle exponential joint action space growth.
ChemCLIP uses contrastive learning to bridge organic and inorganic anticancer compound discovery by enabling knowledge transfer across chemical domains.
LACE mechanism for continual learning that adaptively expands model capacity during training based on loss signal monitoring.
Information-theoretic analysis of safety verification impossibility for self-improving systems balancing bounded risk with unbounded utility.
AMIGO benchmark for evaluating agentic vision-language models on long-horizon multi-image grounding tasks through sequential attribute-focused queries.
GPU-accelerated TensorRT inference pipeline for BERT and GPT-2 with mixed-precision optimization achieving 64.4x CPU speedup.
VeoPlace uses vision-language models for chip floorplanning macro placement by leveraging VLM spatial reasoning abilities to complement learning-based approaches.
HyperP introduces hypersphere parameterization for language model scaling with improved training stability compared to first-order optimizer approaches.
Analysis of why linear probes and sparse autoencoders fail at compositional generalization under superposition, proposing iterative coding alternatives.