CacheTrap achieves 100% targeted attack success on five open-source LLMs by using an efficient search to locate and flip a single bit in the KV cache as a transient trigger, while preserving normal accuracy without the trigger.
Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
A framework detects speaker drift in TTS outputs by computing cosine similarities across speech segments and using LLMs for binary classification, supported by a human-validated synthetic benchmark.
EDITS improves dataset distillation by fusing VLM-generated textual semantics with image features via Global Semantic Query and Local Semantic Awareness modules, then applying Dual Prototype Guidance with an LLM and diffusion model to synthesize compact datasets.
MulFSA combines micro-level firm sentiment, meso-level industry sentiment, and duration-aware smoothing from PLMs/LLMs to extract a daily sentiment index that reduces credit spread forecast errors by 10.25% MAE and 11.94% MAPE on a 1.35M-text Chinese bond corpus.
CommandSwarm uses LoRA-adapted LLMs with safety filtering and deterministic parsing to generate valid behavior trees from text or speech, raising zero-shot BLEU to 0.663 and syntactic validity to 72% on synthetic swarm scenarios.
citing papers explorer
-
CacheTrap: Unveiling a Stealthier Gray-Box Trojan against LLMs
CacheTrap achieves 100% targeted attack success on five open-source LLMs by using an efficient search to locate and flip a single bit in the KV cache as a transient trigger, while preserving normal accuracy without the trigger.
-
A Novel Automatic Framework for Speaker Drift Detection in Synthesized Speech
A framework detects speaker drift in TTS outputs by computing cosine similarities across speech segments and using LLMs for binary classification, supported by a human-validated synthetic benchmark.
-
EDITS: Enhancing Dataset Distillation with Implicit Textual Semantics
EDITS improves dataset distillation by fusing VLM-generated textual semantics with image features via Global Semantic Query and Local Semantic Awareness modules, then applying Dual Prototype Guidance with an LLM and diffusion model to synthesize compact datasets.
-
MulFSA: Multi-level Financial Sentiment Analysis Framework for Bond Market
MulFSA combines micro-level firm sentiment, meso-level industry sentiment, and duration-aware smoothing from PLMs/LLMs to extract a daily sentiment index that reduces credit spread forecast errors by 10.25% MAE and 11.94% MAPE on a 1.35M-text Chinese bond corpus.
-
CommandSwarm: Safety-Aware Natural Language-to-Behavior-Tree Generation for Robotic Swarms
CommandSwarm uses LoRA-adapted LLMs with safety filtering and deterministic parsing to generate valid behavior trees from text or speech, raising zero-shot BLEU to 0.663 and syntactic validity to 72% on synthetic swarm scenarios.