How Architecture and Training Affect TPC Representations Across Experiments
研究通过冻结编码器和探针评估TPC表示在不同实验和探测系统中的可重用性，使用Sparse ResNet和PointNet风格编码器生成固定维度嵌入。
0 citationsRead paper
Selected work
Representative Papers
Solution Space Partitioning for Extremal Set Theory
This work addresses long-standing open problems in extremal set theory, such as Chvátal’s conjecture, by introducing a novel paradigm for customized search space partitioning based on solution construction strategies, replacing conventional domain-agnostic lookahead-based methods. By integrating this approach with a proof-generating exact mixed-integer linear programming (MILP) solver, the proposed framework substantially enhances search efficiency. Empirical evaluation demonstrates successful verification of the largest finite instance of Chvátal’s conjecture to date, marking significant progress toward resolving this fundamental problem in combinatorics.
0 citationsRead paper
A Graph-Native Bitemporal Memory Store for Conversational AI Agents
This work addresses the limitations of conversational AI systems in maintaining persistent memory across sessions, which are often constrained by context window sizes or susceptible to privacy leaks. To overcome these challenges, the authors propose a localized graph-based memory system that integrates a bitemporal model—capturing both valid and transaction time—with a semantic retrieval mechanism. By leveraging immutable memory nodes and automatically constructed semantic edges, the system enables efficient memory association and historical state tracing. Implemented using Neo4j property graphs, HNSW vector indexing, and 1024-dimensional embeddings, it supports point-in-time semantic queries without overwriting past records. Evaluation on the LongMemEval benchmark demonstrates strong performance, achieving 46.7% R@10 for current state retrieval and 80% accuracy on knowledge-update questions, thereby validating the efficacy of the proposed approach.
0 citationsRead paper
Forward versus Backward: Comparing Reasoning Objectives in Direct Preference Optimization
This work addresses the challenge of hallucination in large language models, which often produce plausible yet incorrect answers during reasoning, thereby undermining reliability. The authors propose a novel approach that integrates Direct Preference Optimization (DPO) with Low-Rank Adaptation (LoRA) to simultaneously optimize forward chain-of-thought reasoning and backward verification objectives. Their analysis reveals a fundamental trade-off between these two strategies: forward training improves problem-solving accuracy—raising performance on GSM8K from 83.1% to 86.6%—while backward training substantially reduces the false positive rate, decreasing it from 13.4% to 4.3%. These complementary signals jointly enhance the model’s calibration and output confidence, leading to more trustworthy and self-consistent reasoning.
0 citationsRead paper
When RSA Fails: Exploiting Prime Selection Vulnerabilities in Public Key Cryptography
This work exposes security vulnerabilities in RSA key generation arising from insufficient entropy in embedded device random number generators, leading to flawed prime selection. We identify two practical threats: (i) excessively small prime gaps—enabling Fermat factorization—and (ii) prime reuse or sharing—enabling greatest-common-divisor (GCD) attacks. For the first time, we systematically establish the full causal chain: “hardware entropy deficiency → prime collisions or proximity → scalable private-key recovery.” Through large-scale TLS certificate scanning (covering over 64,000 vulnerable hosts), statistical analysis of prime distributions, entropy evaluation, and multiple cryptanalytic techniques—including Fermat factorization and GCD-based key recovery—we empirically demonstrate that entropy-induced key weaknesses remain widespread. We further propose a lightweight entropy enhancement mechanism and a prime robustness verification scheme, both designed for practical deployment in resource-constrained environments.
0 citationsRead paper
Recent publications
Latest Papers
How Architecture and Training Affect TPC Representations Across Experiments
研究通过冻结编码器和探针评估TPC表示在不同实验和探测系统中的可重用性，使用Sparse ResNet和PointNet风格编码器生成固定维度嵌入。
0 citationsRead paper
Solution Space Partitioning for Extremal Set Theory
This work addresses long-standing open problems in extremal set theory, such as Chvátal’s conjecture, by introducing a novel paradigm for customized search space partitioning based on solution construction strategies, replacing conventional domain-agnostic lookahead-based methods. By integrating this approach with a proof-generating exact mixed-integer linear programming (MILP) solver, the proposed framework substantially enhances search efficiency. Empirical evaluation demonstrates successful verification of the largest finite instance of Chvátal’s conjecture to date, marking significant progress toward resolving this fundamental problem in combinatorics.
0 citationsRead paper
A Graph-Native Bitemporal Memory Store for Conversational AI Agents
This work addresses the limitations of conversational AI systems in maintaining persistent memory across sessions, which are often constrained by context window sizes or susceptible to privacy leaks. To overcome these challenges, the authors propose a localized graph-based memory system that integrates a bitemporal model—capturing both valid and transaction time—with a semantic retrieval mechanism. By leveraging immutable memory nodes and automatically constructed semantic edges, the system enables efficient memory association and historical state tracing. Implemented using Neo4j property graphs, HNSW vector indexing, and 1024-dimensional embeddings, it supports point-in-time semantic queries without overwriting past records. Evaluation on the LongMemEval benchmark demonstrates strong performance, achieving 46.7% R@10 for current state retrieval and 80% accuracy on knowledge-update questions, thereby validating the efficacy of the proposed approach.
0 citationsRead paper
Forward versus Backward: Comparing Reasoning Objectives in Direct Preference Optimization
This work addresses the challenge of hallucination in large language models, which often produce plausible yet incorrect answers during reasoning, thereby undermining reliability. The authors propose a novel approach that integrates Direct Preference Optimization (DPO) with Low-Rank Adaptation (LoRA) to simultaneously optimize forward chain-of-thought reasoning and backward verification objectives. Their analysis reveals a fundamental trade-off between these two strategies: forward training improves problem-solving accuracy—raising performance on GSM8K from 83.1% to 86.6%—while backward training substantially reduces the false positive rate, decreasing it from 13.4% to 4.3%. These complementary signals jointly enhance the model’s calibration and output confidence, leading to more trustworthy and self-consistent reasoning.
0 citationsRead paper
When RSA Fails: Exploiting Prime Selection Vulnerabilities in Public Key Cryptography
This work exposes security vulnerabilities in RSA key generation arising from insufficient entropy in embedded device random number generators, leading to flawed prime selection. We identify two practical threats: (i) excessively small prime gaps—enabling Fermat factorization—and (ii) prime reuse or sharing—enabling greatest-common-divisor (GCD) attacks. For the first time, we systematically establish the full causal chain: “hardware entropy deficiency → prime collisions or proximity → scalable private-key recovery.” Through large-scale TLS certificate scanning (covering over 64,000 vulnerable hosts), statistical analysis of prime distributions, entropy evaluation, and multiple cryptanalytic techniques—including Fermat factorization and GCD-based key recovery—we empirically demonstrate that entropy-induced key weaknesses remain widespread. We further propose a lightweight entropy enhancement mechanism and a prime robustness verification scheme, both designed for practical deployment in resource-constrained environments.
0 citationsRead paper