archive
Every paper Pith has read. Search by title, abstract, or pith.
14513 papers in cs.AI · page 19
-
RL trajectories match real customer paths better than TSP or PNN
Modelling Customer Trajectories with Reinforcement Learning for Practical Retail Insights
-
Accuracy unchanged when latent visual tokens replaced by dummies
What's Holding Back Latent Visual Reasoning?
-
Input flips extend multiplier life under NBTI aging
Building Reliable Arithmetic Multipliers Under NBTI Aging and Process Variations
-
Clean experiences poison reflective LLM agents
OEP: Poisoning Self-Evolving LLM Agents via Locally Correct but Non-Transferable Experiences
-
Dual self-distillation balances privacy and utility in LLMs
It Takes Two: Complementary Self-Distillation for Contextual Integrity in LLMs
-
No memory method works consistently for LLM agents
EvoMemBench: Benchmarking Agent Memory from a Self-Evolving Perspective
-
Geometry-aware coresets lift VLM accuracy in pathology without training
Geometry-Aware Uncertainty Coresets for Robust Visual In-Context Learning in Histopathology
-
Architectural proxy stops unauthorized LLM tool use
Prompts Don't Protect: Architectural Enforcement via MCP Proxy for LLM Tool Access Control
-
AI robotic lab creates graphene and atomically thin transistors
Qumus: Realization of An Embodied AI Quantum Material Experimentalist
5 Piths -
Governed skill libraries boost frozen agents on benchmarks
SkillsVote: Lifecycle Governance of Agent Skills from Collection, Recommendation to Evolution
-
Census simulations diagnose bias in Korean LLMs
Diagnosing Korean-Language LLM Political Bias via Census-Grounded Agent Simulation
-
Graph model beats larger ones on long-range tasks with 1% parameters
Graph Hierarchical Recurrence for Long-Range Generalization
-
Fixed camera network gives robots real-time shared indoor maps
Towards Ubiquitous Mapping and Localization for Dynamic Indoor Environments
-
Hyper-GNN lifts four-top significance to 9.1 sigma
Probing SMEFT Operators through $t\bar{t}t\bar{t}$ Production with Hyper-Graph Neural Networks at the LHC
-
LLMs beat chance on spatial reasoning but stumble on tough calculi
QSTRBench: a New Benchmark to Evaluate the Ability of Language Models to Reason with Qualitative Spatial and Temporal Calculi
4 Piths -
RL fine-tunes LLM to emit reusable solvers 91x cheaper than sampling
Beyond Inference-Time Search: Reinforcement Learning Synthesizes Reusable Solvers
-
AI mirrors user mistakes, lowering advice and performance
The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human-AI Collaboration
-
AI mirrors user mistakes in collaborative ranking tasks
The Hidden Cost of Contextual Sycophancy: an AI Literacy Intervention in Human-AI Collaboration
-
Parameter-free attention matches CSRNet accuracy without extra parameters
Optimising CSRNet with parameter-free attention mechanisms for crowd counting in public transport
-
KV selection per frame and head speeds video diffusion 1.48x
Focused Forcing: Content-Aware Per-Frame KV Selection for Efficient Autoregressive Video Diffusion
-
Framework choice reverses meaning of agent behavior signals
Same Signal, Different Semantics: A Cross-Framework Behavioral Analysis of Software Engineering Agents
-
Feedback steers RL to faster learning and higher peaks
FBOS-RL: Feedback-Driven Bi-Objective Synergistic Reinforcement Learning
-
Causal layer cuts SRE diagnosis time 63%
Causely: A Causal Intelligence Layer for Enterprise AI A Benchmark Study on SRE and Reliability Workflows
-
RAE v2 reaches SOTA gFID 1.06 in 80 epochs on ImageNet
Improved Baselines with Representation Autoencoders
-
Value interpolation expands offline RL action support
ISEP: Implicit Support Expansion for Offline Reinforcement Learning via Stochastic Policy Optimization
-
Wasserstein criterion boosts accuracy of small medical image QA models
Wasserstein Equilibrium Decoding for Reliable Medical Visual Question Answering
-
Prior alignment speeds up re-alignment on re-exposure
Alignment Dynamics in LLM Fine-Tuning
-
Port-Hamiltonian routing shrinks latent space by 4-8% in world models
PH-Dreamer: A Physics-Driven World Model via Port-Hamiltonian Generative Dynamics
-
Self-distillation supplies step-level search signals from own rollouts
SD-Search: On-Policy Hindsight Self-Distillation for Search-Augmented Reasoning
-
Aligning masked EEG views improves cross-dataset transfer
DARE-EEG: A Foundation Model for Mining Dual-Aligned Representation of EEG
-
CommitDistill hits 0.75 retrieval rate from git history at 256-char budget
CommitDistill: A Lightweight Knowledge-Centric Memory Layer for Software Repositories
-
Preference focus cuts device RAG memory 2400 times
From Volume to Value: Preference-Aligned Memory Construction for On-Device RAG
-
Co-training cars and pedestrians cuts collisions 30 percent
Multi-Agent Reinforcement Learning for Safe Autonomous Driving Under Pedestrian Behavioral Uncertainty
-
Prompting methods raise table QA accuracy without training
Efficient Table QA via TableGrid Navigation and Progressive Inference Prompting
-
Shared codebook bridges modalities without full data pairs
CodeBind: Decoupled Representation Learning for Multimodal Alignment with Unified Compositional Codebook
-
MDU unlearns data in masked diffusion models by KL reversal
Machine Unlearning for Masked Diffusion Language Models
-
Privacy RL matches non-private sample bounds in continuous settings
Privacy Preserving Reinforcement Learning with One-Sided Feedback
-
Multi-turn chats in low-resource languages jailbreak LLMs
Multilingual jailbreaking of LLMs using low-resource languages
-
SomaliWeb v1 delivers 303M tokens of cleaned Somali text
SomaliWeb v1: A Quality-Filtered Somali Web Corpus with a Matched Tokenizer and a Public Language-Identification Benchmark
-
Two SAE metrics fail basic reliability checks
Are Sparse Autoencoder Benchmarks Reliable?
-
Memory of precomputed states cuts LLM prefix attention costs
Context Memorization for Efficient Long Context Generation
-
Simplex witness certifies input-dependent VAE encoder
A Simplex Witness Certificate for Constant Collapse in Variational Autoencoders
-
GA-S2S adds k-hop graph structure to raise link prediction 19%
Leveraging Graph Structure in Seq2Seq Models for Knowledge Graph Link Prediction
-
Question routing lifts zero-shot spatial video QA by up to 5%
SPATIOROUTE: Dynamic Prompt Routing for Zero-Shot Spatial Reasoning
-
COCOCO gives conformal sets that obey logic and stay small
Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Models
-
LLM pseudoqueries from table profiles improve dataset search
PIPER: Content-Based Table Search via profiling and LLM-Generated Pseudoqueries
-
RGB cameras build 3D scene graphs for robots as well as depth sensors
RGB-only Active 3D Scene Graph Generation for Indoor Mobile Robots
-
Sensory-bounded reasoning lifts MLLM accuracy on second-order belief tasks
Beyond the Cartesian Illusion: Testing Two-Stage Multi-Modal Theory of Mind under Perceptual Bottlenecks
-
Pairwise preferences boost alignment and diversity in open generation
Pairwise Preference Reward and Group-Based Diversity Enhancement for Superior Open-Ended Generation
-
External cameras boost robot scene recall by up to 79%
Fixed External Cameras as Common Prior Maps for Active 3D Scene Graph Generation