Sieve dynamically schedules MoE experts across GPU and PIM hardware to handle bimodal token distributions, achieving 1.3x to 1.6x gains in throughput and interactivity over static prior PIM systems on three large models.
Mixed citations
Title resolution pending
Mixed citation behavior. Most common role is method (62%).
43
Pith papers citing it
Method
62% of classified citations
citation-role summary
method 5
background 3
citation-polarity summary
representative citing papers
DDIPE poisons LLM agent skills by embedding malicious logic in documentation examples, achieving 11.6-33.5% bypass rates across frameworks while explicit attacks are blocked, with 2.5% evading detection.
PII can be reconstructed from SFT models via prefix attacks, with the new COVA algorithm improving success rates and leakage varying by attacker knowledge and PII type.
GATHER finds topological convergence nodes in a cell-centric knowledge graph to compress multi-gene signals into compact evidence for zero-shot cell-type annotation with a single LLM call.
A co-creation process for inferring and refining personal strivings from computer activity logs yields more representative goals and higher user agency than baselines in a 14-person week-long study.
Orchid benchmark shows requirement ambiguity degrades LLM code generation performance across all models, with advanced models hit hardest, and LLMs rarely detect or resolve the ambiguity themselves.
AccelCIM provides a systematic framework for dataflow exploration in SRAM CIM accelerators using cycle-accurate simulation and post-layout PPA analysis to optimize designs for large language models.
ATI is a tripartite bio-inspired architecture for physical AI that co-designs sensing and inference, shown in a camera prototype to raise accuracy from 53.8% to 88% and cut remote invocations by 43.3%.
Multiple-choice queries synthesized from Hoare triples enable more reliable identification of intended programs than labeled-example supervision in active learning for program disambiguation.
MARINER is a new benchmark dataset and evaluation framework for fine-grained perception and causal reasoning in open-water scenes using 16,629 images across 63 vessel categories, diverse environments, and maritime incidents.
CURE disentangles LLM recommendation circuits into forget-specific, retain-specific, and task-shared modules with tailored update rules to achieve more effective unlearning than weighted baselines.
PrecisionDiff is a differential testing framework that uncovers widespread precision-induced behavioral disagreements in aligned LLMs, including safety-critical jailbreak divergences across precision formats.
Red-Rec uses AI-initiated summaries and low-effort option selection to help users with vague intent explore more broadly and with higher serendipity than user-initiated chat while requiring less typing.
LLM-ODE integrates large language models into genetic programming to guide symbolic search for governing equations of dynamical systems, outperforming classical GP on 91 test cases in efficiency and solution quality.
PhysGen uses video models to learn physics for robots, outperforming baselines by up to 13.8% on Libero and matching specialized models in real-world tasks.
LLMs achieve 81% coherent execution simulation on HumanEval but show mostly random or weak consistency across tests, with frontier models relying on natural language shortcuts instead of true program analysis.
ToolPRM provides fine-grained intra-call process supervision via a new dataset and reward model, outperforming outcome and coarse-grained alternatives on function-calling benchmarks.
IMAGEO-Bench evaluates 10 LLMs on image geolocalization across global street scenes, US POIs, and private images, revealing closed-source model advantages and biases favoring high-resource regions.
SemBid injects LLM-encoded Task, History, and Strategy semantics as tokens into offline bidding trajectories and uses self-attention to outperform numerical-only baselines in performance, constraint satisfaction, and robustness.
NeWTral is a non-linear weight translation framework using MoE routing that reduces average attack success rate from 70% to 13% on unsafe domain adapters across Llama, Mistral, Qwen, and Gemma models up to 72B while retaining 90% knowledge fidelity.
A training-free dual-system framework refines anomaly score ordering on uncertain samples from self-supervised talking head forgery detectors to improve detection performance.
GeoDecider introduces a coarse-to-fine agentic workflow using LLMs for explainable lithology classification from well logs, combining a base classifier, tool-augmented reasoning, and geological refinement to outperform baselines on benchmarks.
TeCoD improves Text-to-SQL execution accuracy by up to 36% over in-context learning and cuts latency 2.2x on matched queries by extracting templates from historical pairs and enforcing them with constrained decoding.
VB-Score shows three major LLMs have severe failures in medical entity recognition and factual consistency, with 13.8% lower performance on chronic conditions affecting older and minority groups, indicating condition-based algorithmic discrimination.
citing papers explorer
-
Sieve: Dynamic Expert-Aware PIM Acceleration for Evolving Mixture-of-Experts Models
Sieve dynamically schedules MoE experts across GPU and PIM hardware to handle bimodal token distributions, achieving 1.3x to 1.6x gains in throughput and interactivity over static prior PIM systems on three large models.
-
Supply-Chain Poisoning Attacks Against LLM Coding Agent Skill Ecosystems
DDIPE poisons LLM agent skills by embedding malicious logic in documentation examples, achieving 11.6-33.5% bypass rates across frameworks while explicit attacks are blocked, with 2.5% evading detection.
-
Reconstruction of Personally Identifiable Information from Supervised Finetuned Models
PII can be reconstructed from SFT models via prefix attacks, with the new COVA algorithm improving success rates and leakage varying by attacker knowledge and PII type.
-
GATHER: Convergence-Centric Hyper-Entity Retrieval for Zero-Shot Cell-Type Annotation
GATHER finds topological convergence nodes in a cell-centric knowledge graph to compress multi-gene signals into compact evidence for zero-shot cell-type annotation with a single LLM call.
-
"What Are You Really Trying to Do?": Co-Creating Life Goals from Everyday Computer Use
A co-creation process for inferring and refining personal strivings from computer activity logs yields more representative goals and higher user agency than baselines in a 14-person week-long study.
-
Assessing the Impact of Requirement Ambiguity on LLM-based Function-Level Code Generation
Orchid benchmark shows requirement ambiguity degrades LLM code generation performance across all models, with advanced models hit hardest, and LLMs rarely detect or resolve the ambiguity themselves.
-
AccelCIM: Systematic Dataflow Exploration for SRAM Compute-in-Memory Accelerator
AccelCIM provides a systematic framework for dataflow exploration in SRAM CIM accelerators using cycle-accurate simulation and post-layout PPA analysis to optimize designs for large language models.
-
[Emerging Ideas] Artificial Tripartite Intelligence: A Bio-Inspired, Sensor-First Architecture for Physical AI
ATI is a tripartite bio-inspired architecture for physical AI that co-designs sensing and inference, shown in a camera prototype to raise accuracy from 53.8% to 88% and cut remote invocations by 43.3%.
-
Choose, Don't Label: Multiple-Choice Query Synthesis for Program Disambiguation
Multiple-choice queries synthesized from Hoare triples enable more reliable identification of intended programs than labeled-example supervision in active learning for program disambiguation.
-
MARINER: A 3E-Driven Benchmark for Fine-Grained Perception and Complex Reasoning in Open-Water Environments
MARINER is a new benchmark dataset and evaluation framework for fine-grained perception and causal reasoning in open-water scenes using 16,629 images across 63 vessel categories, diverse environments, and maritime incidents.
-
CURE:Circuit-Aware Unlearning for LLM-based Recommendation
CURE disentangles LLM recommendation circuits into forget-specific, retain-specific, and task-shared modules with tailored update rules to achieve more effective unlearning than weighted baselines.
-
Hidden Reliability Risks in Large Language Models: Systematic Identification of Precision-Induced Output Disagreements
PrecisionDiff is a differential testing framework that uncovers widespread precision-induced behavioral disagreements in aligned LLMs, including safety-critical jailbreak divergences across precision formats.
-
From Passive Feeds to Guided Discovery: AI-Initiated Interaction for Vague Intent in Content Exploration
Red-Rec uses AI-initiated summaries and low-effort option selection to help users with vague intent explore more broadly and with higher serendipity than user-initiated chat while requiring less typing.
-
LLM-ODE: Data-driven Discovery of Dynamical Systems with Large Language Models
LLM-ODE integrates large language models into genetic programming to guide symbolic search for governing equations of dynamical systems, outperforming classical GP on 91 test cases in efficiency and solution quality.
-
Learning Physics from Pretrained Video Models: A Multimodal Continuous and Sequential World Interaction Models for Robotic Manipulation
PhysGen uses video models to learn physics for robots, outperforming baselines by up to 13.8% on Libero and matching specialized models in real-world tasks.
-
Assessing Coherency and Consistency of Code Execution Reasoning by Large Language Models
LLMs achieve 81% coherent execution simulation on HumanEval but show mostly random or weak consistency across tests, with frontier models relying on natural language shortcuts instead of true program analysis.
-
ToolPRM: Fine-Grained Inference Scaling of Structured Outputs for Function Calling
ToolPRM provides fine-grained intra-call process supervision via a new dataset and reward model, outperforming outcome and coarse-grained alternatives on function-calling benchmarks.
-
From Pixels to Places: A Systematic Benchmark for Evaluating Image Geolocalization Ability in Large Language Models
IMAGEO-Bench evaluates 10 LLMs on image geolocalization across global street scenes, US POIs, and private images, revealing closed-source model advantages and biases favoring high-resource regions.
-
On the Role of Language Representations in Auto-Bidding: Findings and Implications
SemBid injects LLM-encoded Task, History, and Strategy semantics as tokens into offline bidding trajectories and uses self-attention to outperform numerical-only baselines in performance, constraint satisfaction, and robustness.
-
You Snooze, You Lose: Automatic Safety Alignment Restoration through Neural Weight Translation
NeWTral is a non-linear weight translation framework using MoE routing that reduces average attack success rate from 70% to 13% on unsafe domain adapters across Llama, Mistral, Qwen, and Gemma models up to 72B while retaining 90% knowledge fidelity.
-
Enhancing Self-Supervised Talking Head Forgery Detection via a Training-Free Dual-System Framework
A training-free dual-system framework refines anomaly score ordering on uncertain samples from self-supervised talking head forgery detectors to improve detection performance.
-
GeoDecider: A Coarse-to-Fine Agentic Workflow for Explainable Lithology Classification
GeoDecider introduces a coarse-to-fine agentic workflow using LLMs for explainable lithology classification from well logs, combining a base classifier, tool-augmented reasoning, and geological refinement to outperform baselines on benchmarks.
-
Reliable Answers for Recurring Questions: Boosting Text-to-SQL Accuracy with Template Constrained Decoding
TeCoD improves Text-to-SQL execution accuracy by up to 36% over in-context learning and cuts latency 2.2x on matched queries by extracting templates from historical pairs and enforcing them with constrained decoding.
-
Beyond Semantic Similarity: A Component-Wise Evaluation Framework for Medical Question Answering Systems with Health Equity Implications
VB-Score shows three major LLMs have severe failures in medical entity recognition and factual consistency, with 13.8% lower performance on chronic conditions affecting older and minority groups, indicating condition-based algorithmic discrimination.
-
Where to Focus: Query-Modulated Multimodal Keyframe Selection for Long Video Understanding
Q-Gate dynamically routes keyframe selection in long videos via query-modulated gating across visual grounding, global matching, and contextual alignment experts to improve MLLM performance.
-
Mitigating Prompt-Induced Cognitive Biases in General-Purpose AI for Software Engineering
A prompting method that forces GPAI models to state SE best practices before deciding reduces prompt-induced cognitive biases by 51% on average across eight tested biases.
-
An Iterative Test-and-Repair Framework for Competitive Code Generation
A four-stage RL framework that iteratively repairs one code candidate via a code-reading Auditor and a Fixer outperforms larger models and CURE/Specine on competitive programming Pass@1.
-
VC-Soup: Value-Consistency Guided Multi-Value Alignment for Large Language Models
VC-Soup uses a cosine-similarity consistency metric to filter data, trains value-consistent policies, and applies linear merging with Pareto filtering to improve multi-value LLM alignment trade-offs.
-
Analyzing the Presentation, Content, and Utilization of References in LLM-powered Conversational AI Systems
LLM chat systems show large differences in reference quantity and quality, but users rarely click or engage with them.
-
RelianceScope: An Analytical Framework for Examining Students' Reliance on Generative AI Chatbots in Problem Solving
RelianceScope is a new analytical framework that maps AI reliance into nine engagement patterns across help-seeking and response-use, situated in students' prior knowledge and instructional context, validated on programming course logs.
-
BEAR: Towards Beam-Search-Aware Optimization for Recommendation with Large Language Models
BEAR is a cheap token-level top-B regularizer for LLM-based recommendation, but its central claim that this condition is necessary for beam-search survival is incorrect.
-
The Quiet Path from Seemingly Minor Design Errors to Workplace AI Incidents
Empirical analysis of 1,524 AI incident reports shows 83% arise from worker-AI trait misalignments, with 74% of those traceable to developers prioritizing efficiency over precision or personalization.
-
Cheap Expertise: Mapping and Challenging Industry Perspectives in the Expert Data Gig Economy
AI data firms view human expertise as an extractable, low-cost resource to feed AI systems while treating institutional expertise as something needing liberation or reform to fit this model.
-
A Survey on Split Learning for LLM Fine-Tuning: Models, Systems, and Privacy Optimizations
A survey that introduces a unified training pipeline and taxonomizes split learning approaches for LLM fine-tuning across model, system, and privacy dimensions.
-
TACO: Efficient Communication Compression of Intermediate Tensors for Scalable Tensor-Parallel LLM Training
TACO compresses tensor-parallel intermediate tensors with an adaptive FP8 scheme and fused kernels, yielding up to 1.87X throughput gains on GPT and Qwen models with near-lossless accuracy.
-
EcoAssist: Embedding Sustainability into AI-Assisted Frontend Development
EcoAssist embeds energy estimation and optimization into AI-assisted frontend coding, reducing website energy use by 13-16% in benchmarks while preserving developer productivity.
-
How Hyper-Datafication Impacts the Sustainability Costs in Frontier AI
Hyper-datafication—industrialized production of AI training data—creates growing storage-energy, carbon, and labor costs that fall disproportionately on data workers and under-represented languages.
-
Fall into a Pit, Gain in a Wit: Cognitive-Guided Harmful Meme Detection via Misjudgment Risk Pattern Retrieval
PatMD improves harmful meme detection by retrieving misjudgment risk patterns to guide MLLMs, reporting 8.30% average F1 and 7.71% accuracy gains on 6,626 memes across 5 tasks.
-
Reshaping Inclusive Interpersonal Dynamics through Smart Glasses in Mixed-Vision Social Activities
Smart glasses expand independent visual access for BLV participants in mixed-vision groups, supporting inclusive collaboration while sighted participants express uncertainty about adapting their helping behaviors.
-
Taxing Artificial Intelligence
Taxation of AI activities can correct externalities, redistribute costs and gains, and support regulation, though instruments like corporate taxes, consumption taxes, and excises vary in feasibility, measurement challenges, and effects on innovation.
-
Digital Guardians: The Past and The Future of Cyber-Physical Resilience
A survey frames CPS resilience through five themes and illustrates them in connected transportation and medical systems to provide a roadmap for real-world resilience.
-
A Guide to Using Social Media as a Geospatial Lens for Studying Public Opinion and Behavior
Social media data functions as passive geospatial sensing for public opinion and behavior via a structured workflow and case studies on topics like COVID-19 vaccines and urban accessibility.
-
Data-Driven Function Calling Improvements in Large Language Model for Online Financial QA
A pipeline of dataset construction from prior work, AugFC parameter augmentation, and two-step LLM training improves function calling for financial APIs and is running in production.