Institution profile

VISA

Industry researchnorthamerica · us
Official website
Research library25linked papers
Opportunities0open roles
Selected work

Representative Papers

ALERT: Zero-shot LLM Jailbreak Detection via Internal Discrepancy Amplification

Jan 07, 2026arXiv.org

This work addresses the challenge of detecting novel jailbreak attacks against large language models under zero-shot settings, where existing methods often fail to generalize. The authors propose a multi-granularity internal representation discrepancy amplification framework that systematically identifies security-critical components by analyzing model activations at the layer, module, and token levels. By integrating hierarchical, modular, and token-wise feature enhancement mechanisms and employing two complementary classifiers for joint decision-making, the approach effectively uncovers the model’s intrinsic discriminative safety signals. Evaluated on three mainstream safety benchmarks, the method consistently ranks among the top two, achieving average accuracy and F1 score improvements of 10%–40% over the strongest baselines—the first to demonstrate high-precision jailbreak detection in a zero-shot scenario.

2 citationsRead paper

MINT: A Universal Zero-Shot Predictor for Transaction Data

Aug 14, 2026

This study addresses the absence of zero-shot reasoning in foundation models for financial forecasting and the inefficiency of existing LLM-based approaches. We propose MINT, a framework that efficiently aligns trading encoders with LLMs through lightweight embedding injection and instruction tuning. Demonstrating that compact embeddings outperform text serialization, this work establishes a novel multimodal paradigm enabling flexible zero-shot inference. Extensive experiments show that MINT achieves state-of-the-art performance on both in-distribution and out-of-distribution question-answering tasks. Furthermore, it significantly reduces token consumption, inference latency, and memory overhead, thereby realizing efficient and generalizable zero-shot prediction for trading data.

0 citationsRead paper
Recent publications

Latest Papers

MINT: A Universal Zero-Shot Predictor for Transaction Data

Aug 14, 2026

This study addresses the absence of zero-shot reasoning in foundation models for financial forecasting and the inefficiency of existing LLM-based approaches. We propose MINT, a framework that efficiently aligns trading encoders with LLMs through lightweight embedding injection and instruction tuning. Demonstrating that compact embeddings outperform text serialization, this work establishes a novel multimodal paradigm enabling flexible zero-shot inference. Extensive experiments show that MINT achieves state-of-the-art performance on both in-distribution and out-of-distribution question-answering tasks. Furthermore, it significantly reduces token consumption, inference latency, and memory overhead, thereby realizing efficient and generalizable zero-shot prediction for trading data.

0 citationsRead paper

Can LLM Agents Price Competitively? A Dynamic Multi-Attribute Auction Benchmark for Agentic Commerce

Jul 30, 2026

This study investigates the pricing capabilities of large language model (LLM)-driven agents in complex, dynamic markets characterized by hidden customer preferences, real-time competitor responses, and abrupt demand shifts. To this end, we introduce Bazaar, the first dynamic multi-attribute sealed-bid auction benchmark for autonomous commercial agents, which integrates closed-form customer utility functions to balance real-world market complexity with evaluability. Through multi-agent simulations grounded in dynamic game-theoretic modeling, we systematically evaluate eleven state-of-the-art LLMs and uncover a pronounced trade-off between customer acquisition and profit maximization, alongside marked disparities in responsiveness to demand shocks. Notably, even the best-performing LLM agent achieves less than one-third of the ex post optimal profit, highlighting substantial room for improvement in LLM-based autonomous business decision-making.

0 citationsRead paper