19 papers
Aligned Data Can Induce Misalignment via Context Confusion
Yavuz Bakman, Duygu Nur Yaldiz, Baris Askin +4
Large language models (LLMs) are frequently updated for various use cases, where filtering out misaligned training samples is a common practice for preventing post-update misalignm…
Prefill-Free Cross-Family KV Cache Transfer for Heterogeneous Multi-Agent LLMs
Vincent-Daniel Yun, Woosang Lim, Haneul Yoo +3
Recent multi-agent LLM systems increasingly combine heterogeneous models for specialized agent roles. However, text-based communication requires each receiver to prefill shared con…
Output-Aware Rotation for INT2 KV-Cache Quantization
Vincent-Daniel Yun, Woosang Lim, Minsoo Cheong +4
The key-value (KV) cache has become a major memory and bandwidth bottleneck in long-context large language model inference, making ultra-low-bit quantization increasingly important…
Ghosted Layers: Unconstrained Activation Alignment for Recovering Layer-Pruned LLMs
Vincent-Daniel Yun, Junhyuk Jo, Sai Praneeth Karimireddy +1
Layer pruning removes entire Transformer decoder blocks from large language models, but introduces a mismatch between the hidden state received by the next surviving layer and the…
Sampling More, Getting Less: Calibration is the Diversity Bottleneck in LLMs
Amin Banayeeanzade, Qingchuan Yang, Dhruv Tarsadiya +6
Diversity is essential for language-model applications ranging from creative generation to scientific discovery, yet modern LLMs often collapse into a narrow subset of plausible ou…
Robust Multi-Agent LLMs under Byzantine Faults
Haejoon Lee, Vincent-Daniel Yun, Dimitra Panagou +1
Large language model (LLM) agents increasingly collaborate over peer-to-peer networks to improve their reliability. However, these same interactions can also introduce vulnerabilit…
Rethinking Visual Privacy: A Compositional Privacy Risk Framework for Severity Assessment with VLMs
Efthymios Tsaprazlis, Tiantian Feng, Anil Ramakrishna +3
Existing visual privacy benchmarks largely treat privacy as a binary property, labeling images as private or non-private based on visible sensitive content. We argue that privacy i…
Speech Generation Speaker Poisoning: Capability Erasure in Zero-Shot Text-to-Speech
Thanapat Trachu, Thanathai Lertpetchpun, Sai Praneeth Karimireddy +1
Recent zero-shot Text-to-Speech (TTS) systems can clone previously unseen voices from only a few seconds of audio. We formulate Speech Generation Speaker Poisoning (SGSP), a task t…
OpaqueToolsBench: Learning Nuances of Tool Behavior Through Interaction
Skyler Hallinan, Thejas Venkatesh, Xiang Ren +4
Tool-calling is essential for Large Language Model (LLM) agents to complete real-world tasks. While most existing benchmarks assume simple, perfectly documented tools, real-world t…
EPSVec: Efficient and Private Synthetic Data Generation via Dataset Vectors
Amin Banayeeanzade, Qingchuan Yang, Deqing Fu +6
High-quality data is essential for modern machine learning, yet many valuable corpora are sensitive and cannot be freely shared. Synthetic data offers a practical substitute for do…
Sparks of Rationality: Do Reasoning LLMs Align with Human Judgment and Choice?
Ala N. Tak, Amin Banayeeanzade, Anahita Bolourani +5
Large Language Models (LLMs) are increasingly positioned as decision engines for hiring, healthcare, and economic judgment, yet real-world human judgment reflects a balance between…
Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment
Yavuz Bakman, Duygu Nur Yaldiz, Eleni Triantafillou +3
Large Language Models (LLMs) are rarely static and are frequently updated in practice. A growing body of alignment research has shown that models initially deemed ``aligned'' can e…
On the Limits of Momentum in Decentralized and Federated Optimization
Riccardo Zaccone, Sai Praneeth Karimireddy, Carlo Masone
Recent works have explored the use of momentum in local methods to enhance distributed SGD. This is particularly appealing in Federated Learning (FL), where momentum intuitively ap…
DISCO: A Browser-Based Privacy-Preserving Framework for Distributed Collaborative Learning
Julien T. T. Vignoud, Valérian Rousset, Hugo El Guedj +28
Data is often impractical to share for a range of well considered reasons, such as concerns over privacy, intellectual property, and legal constraints. This not only fragments the…
Reject Only Critical Tokens: Pivot-Aware Speculative Decoding
Amir Ziashahabi, Yavuz Faruk Bakman, Duygu Nur Yaldiz +3
Speculative Decoding (SD) ensures that the output matches the target model's distribution exactly. However, we argue that this distribution matching requirement is too stringent an…
Uncertainty as Feature Gaps: Epistemic Uncertainty Quantification of LLMs in Contextual Question-Answering
Yavuz Bakman, Sungmin Kang, Zhiqi Huang +8
Uncertainty Quantification (UQ) research has primarily focused on closed-book factual question answering (QA), while contextual QA remains unexplored, despite its importance in rea…
VoxGuard: Evaluating User and Attribute Privacy in Speech via Membership Inference Attacks
Efthymios Tsaprazlis, Thanathai Lertpetchpun, Tiantian Feng +2
Voice anonymization aims to conceal speaker identity and attributes while preserving intelligibility, but current evaluations rely almost exclusively on Equal Error Rate (EER) that…
The Surprising Effectiveness of Membership Inference with Simple N-Gram Coverage
Skyler Hallinan, Jaehun Jung, Melanie Sclar +7
Membership inference attacks serves as useful tool for fair use of language models, such as detecting potential copyright infringement and auditing data leakage. However, many curr…
TruthTorchLM: A Comprehensive Library for Predicting Truthfulness in LLM Outputs
Duygu Nur Yaldiz, Yavuz Faruk Bakman, Sungmin Kang +9
Generative Large Language Models (LLMs)inevitably produce untruthful responses. Accurately predicting the truthfulness of these outputs is critical, especially in high-stakes setti…