LLMs
Latest breakthroughs in Large Language Models
Articles
Large Models for Battery Prognostics and Health Management: A Review and Future Roadmapβ9
A review of large models for battery prognostics and health management, covering key challenges, methods, and a future roadmap.
Standalone LLM vs. Pre-Specified Agentic Pipeline for Explaining ICU Mortality Predictions: A Feasibilityβ9
Agentic pipelines reduce outcome leakage and improve clinical grounding for ICU mortality explanations, though standalone LLMs better align with SHAP attributio...
TreeGraft: Adaptive Multi-Drafter Grafting for Tree-Based Speculative Decodingβ10
TreeGraft enables adaptive multi-drafter speculative decoding, boosting LLM inference speed by 15.1% across benchmarks.
EduRiskX: A Neuro-Symbolic Framework with F-Logic Reasoning for Early Academic Risk Predictionβ9
EduRiskX combines temporal Transformers with F-Logic reasoning to predict academic risk early, boosting detection rates and interpretability in online learning.
VLM-Based Automatic Multi-Granularity Graph Representation of Building Layouts for Design Informaticsβ7
VLM-based pipeline automatically converts building floorplans into multi-granularity graph representations, achieving 92% node match accuracy for design retriev...
Taming Visual Neglect: A Variational Information Bottleneck Framework for Adaptive Attention in Multimodal Inβ9
New framework reveals when visual context helps or hurts multimodal in-context learning, with adaptive attention boosting accuracy by 4.7%.
RENDER: Controlling Reader-Facing Evidence in LLM Memory Evaluationβ8
Evaluating LLM memory with RENDER: how reader-facing formatsβsummaries, typed records, raw textβimpact accuracy, showing up to 72.6-point variance across models...
Distinguishing Revision and Delayed Elaboration in Incremental Narrative Interpretationβ9
Distinguishing revision from delayed elaboration in incremental narrative interpretation, highlighting their structural differences and roles in processing.
KVBoost: Deviation-Guided Chunk-Level KV Cache Reuse for Efficient LLM Inferenceβ8
KVBoost enables chunk-level KV cache reuse for LLMs, cutting time-to-first-token by 4.49x with 16% gains over prefix caching.
SDAD: Spec-Driven Agentic Development for the AI-Native Software Development Life Cycleβ7
This paper formalizes SDAD, a spec-driven model for AI-native software development, examining its four stages, governance metrics, and impact on the SDLC.