Institution profile

University of Essex

Academic institutioneurope · gb
Official website
Research library162linked papers
Opportunities0open roles
Selected work

Representative Papers

PerspAct: Enhancing LLM Situated Collaboration Skills through Perspective Taking and Active Vision

Nov 11, 2025

Current large language models (LLMs) and multimodal models exhibit limited perspective-taking capabilities in multi-agent collaboration, hindering accurate modeling of subjective agent perceptions and multi-observer environments. To address this, we propose PerspAct—a novel method that integrates active visual exploration with the ReAct reasoning framework for the first time. PerspAct explicitly samples and models diverse agent-centric perspectives, enabling dynamic comprehension of hierarchical perspective complexity in an extended Director task. Built upon multimodal LLMs, it leverages prompt engineering and explicit state representation. We systematically evaluate PerspAct across seven progressively complex scenarios. Experiments demonstrate significant improvements in both coreference resolution and collaborative task accuracy, validating the efficacy of jointly modeling active perception and perspective understanding. Our work establishes a new paradigm for situational awareness in multi-agent settings.

1 citations1 influentialRead paper

MIMO-AFDM Outperforms MIMO-OFDM in the Face of Hardware Impairments

Jan 01, 2026

This study addresses the performance degradation of MIMO multicarrier systems under practical hardware impairments, including phase noise, carrier frequency offset, and nonlinear distortion. The authors systematically analyze the robustness of MIMO-AFDM by deriving a tight upper bound on the bit error rate for maximum-likelihood detection in small-scale systems and a closed-form approximation for linear minimum mean square error detection in large-scale systems, incorporating realistic channel estimation errors. For the first time, they demonstrate that AFDM preserves full diversity gain under both multiplicative and additive hardware impairments and exhibits significantly superior robustness to inter-carrier interference compared to OFDM. Both theoretical analysis and Monte Carlo simulations consistently show that, under identical hardware impairments and across varying mobility scenarios, MIMO-AFDM achieves notably better performance than MIMO-OFDM at moderate to high signal-to-noise ratios.

1 citationsRead paper

Privacy Against Agnostic Inference Attack in Vertical Federated Learning

Feb 10, 2023arXiv.org

This work addresses a novel privacy threat in vertical federated learning (VFL): active parties—equipped with ground-truth labels—perform “agnostic inference attacks” (AIAs) against passive parties’ features, inferring sensitive information without requiring prior knowledge of sample confidence scores. We formally define AIA for the first time and propose a parameter-controllable privacy-preserving mechanism. Methodologically, we design a confidence-augmented logistic regression model, integrating randomized perturbation of passive-party features with a joint privacy-utility optimization framework—preserving model interpretability while enabling rigorous privacy-utility trade-offs. Experiments on benchmark datasets (e.g., CIFAR-10) demonstrate that our approach reduces AIA success rates by over 60%, with less than 2% degradation in model accuracy. This establishes a principled, practical defense against label-leakage-driven inference attacks in VFL settings.

1 citationsRead paper
Recent publications

Latest Papers