Large language model (LLM) agents are increasingly deployed in tool-augmented environments, but their reliance on external inputs makes them highly vulnerable to prompt injection attacks that can hijack task objectives. Existing safety alignment methods rely on static expert trajectories or preference optimization, lim...
Synthetic data are increasingly used as an alternative to sharing sensitive records. However, synthetic data generation does not guarantee privacy, as diffusion models trained or adapted on sensitive data remain susceptible to reconstruction attacks. Moreover, while approaches that use differential privacy (DP), such a...
Bo-Ming Miao, Tao Zhang, Netanel Raviv et al.· 0 citations
As machine learning increasingly relies on public, untrusted data sources, data poisoning attacks, which inject malicious examples into training data to induce misclassification of a chosen target, pose a growing threat. Existing defenses either assume zero ground-truth information about which examples are poisoned, or...
Chaeeun Han, Soodeh Atefi, Yevgeniy Vorobeychik et al.· 0 citations
Bootstrap Flow-Map-Tree (a.k.a BFMT), a computationally efficient sampling framework that enables full tree-path construction from any tree depth using a single function evaluation, drastically reducing computational overhead while providing critical foresight for sequential sampling.
Bing-Lin Ji, Anindya Sarkar, Heng-Chang Lu et al.· 1 citation
It is proved that the resulting sampling framework yields a multi-particle interaction-aware Feynman-Kac corrector that progressively steers the multi-particle system toward a KL-tilted target distribution, facilitating global exploration and preventing mode collapse.
Bing-Lin Ji, Anindya Sarkar, Heng-Chang Lu et al.· 0 citations
AutoDojo is introduced, a generative benchmark built on AgentDojo and AgentDyn that generates IPI adaptively for a given agent and defense, and it is demonstrated that standard static benchmarks often significantly overestimate defense efficacy.
Xinhang Ma, Taoran Li, Chao-Wei Xiao et al.· 5 citations
The Symbolic Guidance Taxonomy (SGT) is introduced, which characterizes a spectrum of autonomy ranging from open-ended natural language reasoning to fully prescribed algorithmic execution, with intermediate levels providing partial pseudocode guidance, and shows that intermediate autonomy levels consistently outperform...
Ben Rachmut, Ning Zhang, Yevgeniy Vorobeychik et al.· Proceedings of the 25th Inte...· 1 citation
Regret-Weighted Payoff Sampling (RWPS), which spends a fixed simulation budget on the payoffs the equilibrium actually depends on and predicts the rest with a model trained on every payoff measured so far, and gives the least exploitable equilibria at the smallest budgets.
Michael Lanier, David Farmer, Yevgeniy Vorobeychik· 0 citations
While diffusion models excel at capturing complex data distributions, scientific discovery often requires steering generation toward specific, uncharacterized regions that maximize a target objective. These high-utility modes frequently reside in low-likelihood tail regions and are only revealed sequentially through in...
Bing-Lin Ji, Anindya Sarkar, Heng-Chang Lu et al.· 0 citations
ROPE is presented, which is anchored in a structural notion of trust, and holds attack success rate to 1.6--2.6 while retaining 82--100% of undefended clean utility, significantly exceeding state-of-the-art system-level defenses in utility while attaining comparable or better security.
Xinhang Ma, Chao-Wei Xiao, William Yeoh et al.· 3 citations· ⚡2
The proposed approach for malicious sentence classification that is both context- and query-aware and outperforms state-of-the-art IPI defense baselines under static attacks, while in the case of adaptive attacks, the AT variants provide significantly higher utility, lower attack success rate, and often both.
