Selected work

Representative Papers

A Session Interaction Framework for The Multiple-Unicast Conjecture

Aug 06, 2026

This study investigates the validity of the multiple-unicast conjecture in undirected networks—namely, whether network coding cannot surpass the throughput limits achievable by routing. To this end, the authors propose a session interaction framework that decomposes verification into two stages: first simplifying the session set via session dominance relations, then applying a session decoupling theorem to the resulting irreducible core to partition it into independent subsets for individual validation. The key contribution lies in establishing, for the first time, an equivalence between the multiple-unicast conjecture and the topological independence of the irreducible core. Furthermore, the work introduces geometric simplification conditions and a high-cost cut-set criterion, yielding a computable and iterative verification mechanism applicable to arbitrary undirected networks, thereby advancing the intersection of network information theory and computational complexity theory.

0 citationsRead paper

Disentanglement in Difference: Directly Learning Semantically Disentangled Representations by Maximizing Inter-Factor Differences

Feb 05, 2025

In disentangled representation learning, statistical independence does not guarantee semantic irrelevance, rendering conventional independence-based methods incapable of ensuring semantic disentanglement. This work identifies this fundamental inconsistency and proposes a novel “difference-driven disentanglement” paradigm: it abandons the latent-variable independence assumption and instead explicitly models the intrinsic semantic distinctions among factors. Specifically, we design a difference encoder to capture discriminative semantic features across factors and introduce a cross-dimensional contrastive loss to achieve explicit, semantic-level disentanglement in a fully unsupervised manner. Evaluated on dSprites and 3DShapes, our method consistently outperforms state-of-the-art disentanglement models across multiple standard metrics—including DCI, SAP, and MIG—demonstrating both the effectiveness and generalizability of semantic difference modeling for disentanglement.

0 citationsRead paper

Convolution-Based Converter : A Weak-Prior Approach For Modeling Stochastic Processes Based On Conditional Density Estimation

Feb 05, 2025

This paper addresses the limited generalizability of stochastic process modeling caused by strong parametric assumptions—such as Markovianity and Gaussianity. To overcome this, we propose the Convolutional Bridge Constructor (CBC), a weak-prior, data-driven framework for conditional density estimation. CBC eschews predefined dynamical structures and distributional forms, instead learning the target variable’s probability distribution and constraint-satisfying expected trajectories implicitly from data via observation-conditioned embedding and end-to-end trajectory generation. Its core innovation is the first nonparametric, convolutional modeling paradigm explicitly designed for stochastic processes. In comprehensive benchmark evaluations, CBC consistently outperforms Markov chains, Gaussian processes, and state-of-the-art deep generative models in prediction accuracy, probabilistic calibration, and uncertainty quantification. Notably, it demonstrates significantly enhanced robustness and adaptability under out-of-distribution and non-stationary conditions.

0 citationsRead paper

Recent publications

Latest Papers

A Session Interaction Framework for The Multiple-Unicast Conjecture

Aug 06, 2026

This study investigates the validity of the multiple-unicast conjecture in undirected networks—namely, whether network coding cannot surpass the throughput limits achievable by routing. To this end, the authors propose a session interaction framework that decomposes verification into two stages: first simplifying the session set via session dominance relations, then applying a session decoupling theorem to the resulting irreducible core to partition it into independent subsets for individual validation. The key contribution lies in establishing, for the first time, an equivalence between the multiple-unicast conjecture and the topological independence of the irreducible core. Furthermore, the work introduces geometric simplification conditions and a high-cost cut-set criterion, yielding a computable and iterative verification mechanism applicable to arbitrary undirected networks, thereby advancing the intersection of network information theory and computational complexity theory.

0 citationsRead paper

Disentanglement in Difference: Directly Learning Semantically Disentangled Representations by Maximizing Inter-Factor Differences

Feb 05, 2025

In disentangled representation learning, statistical independence does not guarantee semantic irrelevance, rendering conventional independence-based methods incapable of ensuring semantic disentanglement. This work identifies this fundamental inconsistency and proposes a novel “difference-driven disentanglement” paradigm: it abandons the latent-variable independence assumption and instead explicitly models the intrinsic semantic distinctions among factors. Specifically, we design a difference encoder to capture discriminative semantic features across factors and introduce a cross-dimensional contrastive loss to achieve explicit, semantic-level disentanglement in a fully unsupervised manner. Evaluated on dSprites and 3DShapes, our method consistently outperforms state-of-the-art disentanglement models across multiple standard metrics—including DCI, SAP, and MIG—demonstrating both the effectiveness and generalizability of semantic difference modeling for disentanglement.

0 citationsRead paper

Convolution-Based Converter : A Weak-Prior Approach For Modeling Stochastic Processes Based On Conditional Density Estimation

Feb 05, 2025

This paper addresses the limited generalizability of stochastic process modeling caused by strong parametric assumptions—such as Markovianity and Gaussianity. To overcome this, we propose the Convolutional Bridge Constructor (CBC), a weak-prior, data-driven framework for conditional density estimation. CBC eschews predefined dynamical structures and distributional forms, instead learning the target variable’s probability distribution and constraint-satisfying expected trajectories implicitly from data via observation-conditioned embedding and end-to-end trajectory generation. Its core innovation is the first nonparametric, convolutional modeling paradigm explicitly designed for stochastic processes. In comprehensive benchmark evaluations, CBC consistently outperforms Markov chains, Gaussian processes, and state-of-the-art deep generative models in prediction accuracy, probabilistic calibration, and uncertainty quantification. Notably, it demonstrates significantly enhanced robustness and adaptability under out-of-distribution and non-stationary conditions.

0 citationsRead paper