Institution profile

Brandeis University

Academic institutionnorthamerica · us
Official website
Research library57linked papers
Opportunities0open roles
Selected work

Representative Papers

Proof of Time: A Benchmark for Evaluating Scientific Idea Judgments

Jan 12, 2026

Existing approaches struggle to scalably evaluate the quality of large language models’ judgments on scientific ideas. This work proposes the Prediction-over-Time (PoT) benchmark framework, which constructs offline sandboxes via temporal slicing and predicts subsequent observable signals—such as citation counts or shifts in research agendas—based solely on evidence available up to a frozen cutoff date, thereby enabling verifiable evaluation without extensive expert annotations. Integrating tool-augmented agents, prompt ablation, and budget-scaling strategies, PoT is validated across over 30,000 instances spanning four scientific domains. Results show that increasing interaction budgets generally improves performance over non-agent baselines, while the efficacy of tool use is highly task-dependent. The framework further supports analyses of human–agent judgment alignment and enables controlled evaluations of agent-based scientific reviewing.

1 citationsRead paper

Dynamic Epistemic Friction in Dialogue

Jun 12, 2025

This work addresses the cognitive resistance to belief updating in human–agent collaborative dialogue, introducing the concept of “dynamic cognitive friction”—a systematic resistance to integrating novel, conflicting, or ambiguous external evidence with preexisting beliefs. Methodologically, it formalizes cognitive friction for the first time within a dynamic epistemic logic (DEL) framework as a nontrivial belief revision process, constructs a quantifiable friction model grounded in belief alignment metrics, and integrates formal belief revision theory with empirical dialogue analysis. Evaluated in embodied collaborative tasks, the model successfully predicts belief update trajectories observed in real human–agent dialogues. It significantly improves modeling accuracy and explanatory power for complex interactive cognitive dynamics. The approach provides a novel theoretical framework and quantitative foundation for interpretable human–agent collaboration.

1 citationsRead paper
Recent publications

Latest Papers

Exploring Privacy Leakage and Data Disclosure Violations in the MacOS Application Ecosystem

Aug 05, 2026

This study addresses the widespread inconsistencies in privacy disclosures and insufficient transparency regarding data collection practices within the macOS desktop application ecosystem, which lacks cohesive mechanisms and systematic investigation. To bridge this gap, we present NutriScan, a novel framework that integrates static and dynamic analysis to comprehensively track API invocations, outbound data transmissions, and alignment with stated privacy policies. Applying NutriScan to 1,000 popular macOS applications reveals that 85% fail to disclose their use of sensitive APIs, while 49.7% transmit user data to third parties—12.5% of which provide no disclosure whatsoever. Further analysis indicates that such data is commonly leveraged for user profiling and device fingerprinting. Based on these findings, we propose concrete mitigations to enhance developer disclosure practices and strengthen App Store review protocols.

0 citationsRead paper

How Formerly Incarcerated People Envision Technologies for Prison Parole

Jul 17, 2026

This study addresses the lack of involvement from communities impacted by the criminal legal system in the design of parole calculation tools, which often fail to reflect their actual needs. Drawing on in-depth interviews with 31 formerly incarcerated individuals and employing participatory envisioning methods, the research proposes a design perspective centered on “strategic agency,” emphasizing that technology should empower individuals and their families to navigate and contest existing power structures in pursuit of freedom, rather than reinforcing surveillance. Key design directions identified include cultural responsiveness, visualization of personal transformation, and recognition of familial labor. These insights reveal authentic expectations for parole-related technologies and offer a new paradigm for developing human-centered, agency-supportive parole assistance systems.

0 citationsRead paper