Institution profile

University of Athens

Academic institutioneurope · gr
Official website
Research library26linked papers
Opportunities0open roles
Selected work

Representative Papers

A sampling Lovász Local Lemma

Sep 16, 2026

本文提出了一种近似均匀采样器，用于解决满足特定条件的约束满足问题，通过调用一种高效的近似计数算法，在多项式时间内生成接近均匀分布的解。

0 citationsRead paper

ZKP Security Tools and Verification: Coverage, Effectiveness, Adoption, and Challenges

Jul 26, 2026

Zero-knowledge proof (ZKP) systems are highly complex to implement, where subtle errors can compromise their security guarantees, yet the effectiveness and coverage of existing security tools in real-world settings remain unclear. This work presents the first systematic evaluation of ZKP security tools, integrating vulnerability benchmarking, formal verification analysis, and a large-scale survey of practitioners across mainstream ecosystems such as Circom-based DSLs and zkVMs. The study reveals that while current tools detect 45.7% of vulnerabilities on isolated targets, their performance drops sharply to 19.6% in full projects; formal verification efforts predominantly focus on constraint correctness; and developers heavily rely on manual processes and widely adopt large language models (LLMs). These findings underscore an urgent need for tools offering low integration overhead and clearer security assurances. The paper contributes the first empirical evaluation framework for ZKP security tools, a comprehensive review of formal verification systems, and actionable practitioner insights.

0 citationsRead paper

Neural Collapse by Design: Learning Class Prototypes on the Hypersphere

May 19, 2026

This work addresses the practical difficulty of achieving Neural Collapse—the theoretically optimal state in supervised classification—by proposing a unified hyperspherical prototype contrastive framework that integrates cross-entropy and supervised contrastive learning. The approach employs normalized losses (NTCE and NONL) to enhance negative sample utilization and decouple alignment from uniformity, while theoretically demonstrating that supervised contrastive learning inherently yields an optimal classifier, obviating the need for linear probing. By unifying two dominant paradigms under a prototype contrastive perspective and using class-mean embeddings as classifier weights, the method closely approximates the geometric structure of Neural Collapse. It outperforms standard cross-entropy on four benchmarks including ImageNet-1K, with over 95% of metrics approaching theoretical limits, achieves collapse characteristics within just 7.5% of training iterations, improves transfer learning performance by 5.5% on average, gains up to 8.7% under severe class imbalance, and effectively reduces mCE on ImageNet-C.

0 citationsRead paper
Recent publications

Latest Papers

A sampling Lovász Local Lemma

Sep 16, 2026

本文提出了一种近似均匀采样器，用于解决满足特定条件的约束满足问题，通过调用一种高效的近似计数算法，在多项式时间内生成接近均匀分布的解。

0 citationsRead paper

ZKP Security Tools and Verification: Coverage, Effectiveness, Adoption, and Challenges

Jul 26, 2026

Zero-knowledge proof (ZKP) systems are highly complex to implement, where subtle errors can compromise their security guarantees, yet the effectiveness and coverage of existing security tools in real-world settings remain unclear. This work presents the first systematic evaluation of ZKP security tools, integrating vulnerability benchmarking, formal verification analysis, and a large-scale survey of practitioners across mainstream ecosystems such as Circom-based DSLs and zkVMs. The study reveals that while current tools detect 45.7% of vulnerabilities on isolated targets, their performance drops sharply to 19.6% in full projects; formal verification efforts predominantly focus on constraint correctness; and developers heavily rely on manual processes and widely adopt large language models (LLMs). These findings underscore an urgent need for tools offering low integration overhead and clearer security assurances. The paper contributes the first empirical evaluation framework for ZKP security tools, a comprehensive review of formal verification systems, and actionable practitioner insights.

0 citationsRead paper

Neural Collapse by Design: Learning Class Prototypes on the Hypersphere

May 19, 2026

This work addresses the practical difficulty of achieving Neural Collapse—the theoretically optimal state in supervised classification—by proposing a unified hyperspherical prototype contrastive framework that integrates cross-entropy and supervised contrastive learning. The approach employs normalized losses (NTCE and NONL) to enhance negative sample utilization and decouple alignment from uniformity, while theoretically demonstrating that supervised contrastive learning inherently yields an optimal classifier, obviating the need for linear probing. By unifying two dominant paradigms under a prototype contrastive perspective and using class-mean embeddings as classifier weights, the method closely approximates the geometric structure of Neural Collapse. It outperforms standard cross-entropy on four benchmarks including ImageNet-1K, with over 95% of metrics approaching theoretical limits, achieves collapse characteristics within just 7.5% of training iterations, improves transfer learning performance by 5.5% on average, gains up to 8.7% under severe class imbalance, and effectively reduces mCE on ImageNet-C.

0 citationsRead paper