arXiv subjects

Zhuan Shi

Publications and source records attributed to Zhuan Shi.

15 recordsLinked to original sources

arXiv2026

CABAL: Multi-Agent Simulacra for Tracing the Effects of Collusive Bidding in Peer Review

Recent reports during the AAAI-27 review cycle highlight the risk of reviewers coordinating bids for reciprocal assignment advantage. Prior work treats bidding, reviewer assignment, and review manipulation as separate stages, leaving the lifecycle effects of collusive bidding unclear. Real-world analysis is further constrained by typically unobservable collusive intent and the lack of counterfactuals for the same conference. Motivated by this gap, we introduce \alg, an end-to-end multi-agent simulacra framework for studying reviewer assignment integrity by holding the conference environment fixed and configuring LLM-driven reviewer agents with honest or collusive policies. We further develop an affinity-guided collusive bidding strategy that uses mutual reviewer-paper affinities to construct collusion rings and select target papers, producing expertise-consistent rather than arbitrarily targeted attacks. Controlled experiments show that collusive bidding more than doubles target-paper capture and that assigned colluders score target papers about two points higher than honest co-reviewers, while conference-wide effects remain comparatively modest. Evaluated bid-phase detectors provide only limited evidence of collusion: in a fixed-triplet detector stress test, native positive-bid graphs are confounded by benign affinity, while a Very-High-only diagnostic view enables precise but low-coverage local recovery.

cs.AI↗

arXiv2026

TEA: Text Encoder Alignment for Robust Concept Erasure in Text-to-Image Models

Text-to-image diffusion models can be misused to generate harmful content through adversarial or paraphrased prompts that bypass built-in safety mechanisms. Existing concept erasure methods often suffer from limited robustness against adversarial prompts, degradation of benign generation quality, or reliance on inference-time interventions that introduce persistent computational overhead. To address these limitations, we formulate concept erasure as a domain alignment problem in the text representation space. We propose a lightweight Text Encoder Alignment framework (TEA) that fine-tunes only the text encoder while keeping the generative backbone fully frozen. Given concept--anchor prompt pairs, our method trains a discriminator to distinguish token-level representations of concept-containing prompts from those of safe anchor prompts, while updating the text encoder to make these representations indistinguishable. TEA introduces zero inference-time overhead and requires only a small number of fine-tuning steps, making it highly efficient to deploy at scale. Despite this efficiency, TEA achieves state-of-the-art erasure robustness against black-box and white-box adversarial attacks on Stable Diffusion v1.4, while preserving generation quality on benign prompts. Furthermore, TEA is model-agnostic and achieves the lowest attack success rate on Stable Diffusion v3.5, extending concept erasure to a Rectified Flow Transformer architecture with T5 conditioning where prior methods remain largely unexplored. Code is available at \href{https://github.com/alirezafarashah/TEA.git}{https://github.com/alirezafarashah/TEA.git}

cs.CV↗

arXiv2026

IP Protection in the Era of Visual Generative AI: A Survey

The rapid evolution of visual generative AI has introduced a wide range of intellectual property risks, spanning the unauthorized learning, reproduction, extraction, misuse, and redistribution of protected data and model assets. To address these risks, a growing body of technical defenses has been proposed. However, existing surveys typically organize this literature by lifecycle stage or technical mechanism, which can obscure the protective intent of different methods. This survey presents a two-dimensional taxonomy for IP protection in visual generative models. The primary axis is a Control Logic View, which classifies methods into Information Exposure Control, Generative Behavior Constraint, and Attribution & Accountability according to the risk variable they regulate. The secondary axis distinguishes Data IP from Model IP as cross-cutting asset dimensions. Under this framework, we systematically review protection methods, align evaluation protocols with protection objectives, and discuss open challenges including proactive model-level safeguards, standardized evaluation, robustness against adaptive attacks, and explainable evidence. This survey aims to offer a principled, systematic, and easy-to-follow overview for both new and experienced researchers in visual generative AI IP protection.

cs.CV↗

arXiv2026

How and What to Imagine? Visual Thinking in Unified Multimodal Models for Cross-View Spatial Reasoning

Cross-view spatial reasoning remains a weak spot for vision-language models (VLMs): they often reason in language and lose the fine-grained geometry needed for the task. Thinking with images aims to address this by generating an intermediate thinking image, but recent work shows that models often ignore the visual evidence in these traces. We therefore ask how to make visual thinking matter, and what kind of visual thinking works best. We study these questions in unified multimodal models (UMMs), which natively support interleaved image-text generation. For the first question, we propose View Dropout (VDrop), a training-time intervention that hides parts of one input view from the answer span while keeping them visible to the thinking-image tokens. This encourages the model to use the thinking image when answering, instead of relying only on the input views. Once the thinking image is used for answer prediction, we study which type of visual thinking is most effective. We frame this as a learnability-informativeness tradeoff and compare three thinking-image variants: top-down, panoramic, and point-matching renderings. Trained on synthetic scenes and evaluated on five real-world out-of-domain benchmarks, panoramic visual thinking with VDrop is the only configuration that is both informative and learnable, and it achieves the best out-of-domain generalization.

cs.CV↗

arXiv2026

Neighbor-Aware Localized Concept Erasure in Text-to-Image Diffusion Models

Concept erasure in text-to-image diffusion models seeks to remove undesired concepts while preserving overall generative capability. Localized erasure methods aim to restrict edits to the spatial region occupied by the target concept. However, we observe that suppressing a concept can unintentionally weaken semantically related neighbor concepts, reducing fidelity in fine-grained domains. We propose Neighbor-Aware Localized Concept Erasure (NLCE), a training-free framework designed to better preserve neighboring concepts while removing target concepts. It operates in three stages: (1) a spectrally-weighted embedding modulation that attenuates target concept directions while stabilizing neighbor concept representations, (2) an attention-guided spatial gate that identifies regions exhibiting residual concept activation, and (3) a spatially-gated hard erasure that eliminates remaining traces only where necessary. This neighbor-aware pipeline enables localized concept removal while maintaining the surrounding concept neighborhood structure. Experiments on fine-grained datasets (Oxford Flowers, Stanford Dogs) show that our method effectively removes target concepts while better preserving closely related categories. Additional results on celebrity identity, explicit content and artistic style demonstrate robustness and generalization to broader erasure scenarios.

cs.CV↗

arXiv2026

Multilingual Amnesia: On the Transferability of Unlearning in Multilingual LLMs

As multilingual large language models become more widely used, ensuring their safety and fairness across diverse linguistic contexts presents unique challenges. While existing research on machine unlearning has primarily focused on monolingual settings, typically English, multilingual environments introduce additional complexities due to cross-lingual knowledge transfer and biases embedded in both pretraining and fine-tuning data. In this work, we study multilingual unlearning using the Aya-Expanse 8B model under two settings: (1) data unlearning and (2) concept unlearning. We extend benchmarks for factual knowledge and stereotypes to ten languages through translation: English, French, Arabic, Japanese, Russian, Farsi, Korean, Hindi, Hebrew, and Indonesian. These languages span five language families and a wide range of resource levels. Our experiments show that unlearning in high-resource languages is generally more stable, with asymmetric transfer effects observed between typologically related languages. Furthermore, our analysis of linguistic distances indicates that syntactic similarity is the strongest predictor of cross-lingual unlearning behavior.

cs.CL↗

arXiv2025

CopyJudge: Automated Copyright Infringement Identification and Mitigation in Text-to-Image Diffusion Models

Assessing whether AI-generated images are substantially similar to source works is a crucial step in resolving copyright disputes. In this paper, we propose CopyJudge, a novel automated infringement identification framework that leverages large vision-language models (LVLMs) to simulate practical court processes for determining substantial similarity between copyrighted images and those generated by text-to-image diffusion models. Specifically, we employ an abstraction-filtration-comparison test framework based on the multi-LVLM debate to assess the likelihood of infringement and provide detailed judgment rationales. Based on these judgments, we further introduce a general LVLM-based mitigation strategy that automatically optimizes infringing prompts by avoiding sensitive expressions while preserving the non-infringing content. Furthermore, assuming the input noise is controllable, our approach can be enhanced by iteratively exploring non-infringing noise vectors within the diffusion latent space, even without modifying the original prompts. Experimental results show that our automated identification method achieves comparable state-of-the-art performance, while offering superior generalization and interpretability across various forms of infringement, and that our mitigation method more effectively mitigates memorization and IP infringement with a high degree of alignment to the original non-infringing expressions.

cs.CV↗

arXiv2025

Reviving Your MNEME: Predicting The Side Effects of LLM Unlearning and Fine-Tuning via Sparse Model Diffing

Large language models (LLMs) are frequently fine-tuned or unlearned to adapt to new tasks or eliminate undesirable behaviors. While existing evaluation methods assess performance after such interventions, there remains no general approach for detecting unintended side effects, such as unlearning biology content degrading performance on chemistry tasks, particularly when these effects are unpredictable or emergent. To address this issue, we introduce MNEME, Model diffiNg for Evaluating Mechanistic Effects, a lightweight framework for identifying these side effects using sparse model diffing. MNEME compares base and fine-tuned models on task-agnostic data (for example, The Pile, LMSYS-Chat-1M) without access to fine-tuning data to isolate behavioral shifts. Applied to five LLMs across three scenarios: WMDP knowledge unlearning, emergent misalignment, and benign fine-tuning, MNEME achieves up to 95 percent accuracy in predicting side effects, aligning with known benchmarks and requiring no custom heuristics. Furthermore, we show that retraining on high-activation samples can partially reverse these effects. Our results demonstrate that sparse probing and diffing offer a scalable and automated lens into fine-tuning-induced model changes, providing practical tools for understanding and managing LLM behavior.

cs.CL↗

arXiv2025

FedCDC: A Collaborative Framework for Data Consumers in Federated Learning Market

Federated learning (FL) allows machine learning models to be trained on distributed datasets without directly accessing local data. In FL markets, numerous Data Consumers compete to recruit Data Owners for their respective training tasks, but budget constraints and competition can prevent them from securing sufficient data. While existing solutions focus on optimizing one-to-one matching between Data Owners and Data Consumers, we propose \methodname{}, a novel framework that facilitates collaborative recruitment and training for Data Consumers with similar tasks. Specifically, \methodname{} detects shared subtasks among multiple Data Consumers and coordinates the joint training of submodels specialized for these subtasks. Then, through ensemble distillation, these submodels are integrated into each Data Consumer global model. Experimental evaluations on three benchmark datasets demonstrate that restricting Data Consumers access to Data Owners significantly degrades model performance; however, by incorporating \methodname{}, this performance loss is effectively mitigated, resulting in substantial accuracy gains for all participating Data Consumers.

cs.DC↗

arXiv2025

RLCP: A Reinforcement Learning-based Copyright Protection Method for Text-to-Image Diffusion Model

The increasing sophistication of text-to-image generative models has led to complex challenges in defining and enforcing copyright infringement criteria and protection. Existing methods, such as watermarking and dataset deduplication, fail to provide comprehensive solutions due to the lack of standardized metrics and the inherent complexity of addressing copyright infringement in diffusion models. To deal with these challenges, we propose a Reinforcement Learning-based Copyright Protection(RLCP) method for Text-to-Image Diffusion Model, which minimizes the generation of copyright-infringing content while maintaining the quality of the model-generated dataset. Our approach begins with the introduction of a novel copyright metric grounded in copyright law and court precedents on infringement. We then utilize the Denoising Diffusion Policy Optimization (DDPO) framework to guide the model through a multi-step decision-making process, optimizing it using a reward function that incorporates our proposed copyright metric. Additionally, we employ KL divergence as a regularization term to mitigate some failure modes and stabilize RL fine-tuning. Experiments conducted on 3 mixed datasets of copyright and non-copyright images demonstrate that our approach significantly reduces copyright infringement risk while maintaining image quality.

cs.CY↗

arXiv2024

Copyright-Aware Incentive Scheme for Generative Art Models Using Hierarchical Reinforcement Learning

Generative art using Diffusion models has achieved remarkable performance in image generation and text-to-image tasks. However, the increasing demand for training data in generative art raises significant concerns about copyright infringement, as models can produce images highly similar to copyrighted works. Existing solutions attempt to mitigate this by perturbing Diffusion models to reduce the likelihood of generating such images, but this often compromises model performance. Another approach focuses on economically compensating data holders for their contributions, yet it fails to address copyright loss adequately. Our approach begin with the introduction of a novel copyright metric grounded in copyright law and court precedents on infringement. We then employ the TRAK method to estimate the contribution of data holders. To accommodate the continuous data collection process, we divide the training into multiple rounds. Finally, We designed a hierarchical budget allocation method based on reinforcement learning to determine the budget for each round and the remuneration of the data holder based on the data holder's contribution and copyright loss in each round. Extensive experiments across three datasets show that our method outperforms all eight benchmarks, demonstrating its effectiveness in optimizing budget distribution in a copyright-aware manner. To the best of our knowledge, this is the first technical work that introduces to incentive contributors and protect their copyrights by compensating them.

cs.LG↗

arXiv2024

FedSSA: Semantic Similarity-based Aggregation for Efficient Model-Heterogeneous Personalized Federated Learning

Federated learning (FL) is a privacy-preserving collaboratively machine learning paradigm. Traditional FL requires all data owners (a.k.a. FL clients) to train the same local model. This design is not well-suited for scenarios involving data and/or system heterogeneity. Model-Heterogeneous Personalized FL (MHPFL) has emerged to address this challenge. Existing MHPFL approaches often rely on a public dataset with the same nature as the learning task, or incur high computation and communication costs. To address these limitations, we propose the Federated Semantic Similarity Aggregation (FedSSA) approach for supervised classification tasks, which splits each client's model into a heterogeneous (structure-different) feature extractor and a homogeneous (structure-same) classification header. It performs local-to-global knowledge transfer via semantic similarity-based header parameter aggregation. In addition, global-to-local knowledge transfer is achieved via an adaptive parameter stabilization strategy which fuses the seen-class parameters of historical local headers with that of the latest global header for each client. FedSSA does not rely on public datasets, while only requiring partial header parameter transmission to save costs. Theoretical analysis proves the convergence of FedSSA. Extensive experiments present that FedSSA achieves up to 3.62% higher accuracy, 15.54 times higher communication efficiency, and 15.52 times higher computational efficiency compared to 7 state-of-the-art MHPFL baselines.

cs.LG↗

arXiv2023

Towards AI-Empowered Crowdsourcing

Crowdsourcing, in which human intelligence and productivity is dynamically mobilized to tackle tasks too complex for automation alone to handle, has grown to be an important research topic and inspired new businesses (e.g., Uber, Airbnb). Over the years, crowdsourcing has morphed from providing a platform where workers and tasks can be matched up manually into one which leverages data-driven algorithmic management approaches powered by artificial intelligence (AI) to achieve increasingly sophisticated optimization objectives. In this paper, we provide a survey presenting a unique systematic overview on how AI can empower crowdsourcing to improve its efficiency - which we refer to as AI-Empowered Crowdsourcing(AIEC). We propose a taxonomy which divides AIEC into three major areas: 1) task delegation, 2) motivating workers, and 3) quality control, focusing on the major objectives which need to be accomplished. We discuss the limitations and insights, and curate the challenges of doing research in each of these areas to highlight promising future research directions.

cs.CY↗

arXiv2023

FedGH: Heterogeneous Federated Learning with Generalized Global Header

Federated learning (FL) is an emerging machine learning paradigm that allows multiple parties to train a shared model collaboratively in a privacy-preserving manner. Existing horizontal FL methods generally assume that the FL server and clients hold the same model structure. However, due to system heterogeneity and the need for personalization, enabling clients to hold models with diverse structures has become an important direction. Existing model-heterogeneous FL approaches often require publicly available datasets and incur high communication and/or computational costs, which limit their performances. To address these limitations, we propose a simple but effective Federated Global prediction Header (FedGH) approach. It is a communication and computation-efficient model-heterogeneous FL framework which trains a shared generalized global prediction header with representations extracted by heterogeneous extractors for clients' models at the FL server. The trained generalized global prediction header learns from different clients. The acquired global knowledge is then transferred to clients to substitute each client's local prediction header. We derive the non-convex convergence rate of FedGH. Extensive experiments on two real-world datasets demonstrate that FedGH achieves significantly more advantageous performance in both model-homogeneous and -heterogeneous FL scenarios compared to seven state-of-the-art personalized FL models, beating the best-performing baseline by up to 8.87% (for model-homogeneous FL) and 1.83% (for model-heterogeneous FL) in terms of average test accuracy, while saving up to 85.53% of communication overhead.

cs.LG↗

arXiv2023

Fairness-Aware Client Selection for Federated Learning

Federated learning (FL) has enabled multiple data owners (a.k.a. FL clients) to train machine learning models collaboratively without revealing private data. Since the FL server can only engage a limited number of clients in each training round, FL client selection has become an important research problem. Existing approaches generally focus on either enhancing FL model performance or enhancing the fair treatment of FL clients. The problem of balancing performance and fairness considerations when selecting FL clients remains open. To address this problem, we propose the Fairness-aware Federated Client Selection (FairFedCS) approach. Based on Lyapunov optimization, it dynamically adjusts FL clients' selection probabilities by jointly considering their reputations, times of participation in FL tasks and contributions to the resulting model performance. By not using threshold-based reputation filtering, it provides FL clients with opportunities to redeem their reputations after a perceived poor performance, thereby further enhancing fair client treatment. Extensive experiments based on real-world multimedia datasets show that FairFedCS achieves 19.6% higher fairness and 0.73% higher test accuracy on average than the best-performing state-of-the-art approach.

cs.LG↗