Keyue Jiang

Xiaohang Tang, Keyue Jiang, Che Liu et al.

Reinforcement learning (RL) can be used to improve the policy (denoiser) of diffusion large language models (dLLMs), while being hindered by the intractability of the policy likelihood. A dominant and efficient family of methods replaces the likelihood in standard RL with its evidence lower bound (ELBO), estimated from randomly masked sequences. Despite being well aligned with pre-training, these approaches introduce bias through training--inference mismatch by using the ELBO as a likelihood surrogate, which can degrade performance. In this work, we propose Guided Denoiser Self-Distillation (GDSD) to directly distill the denoiser of dLLMs from an advantage-guided self-teacher, derived from the closed-form optimum of reverse-KL regularized RL. GDSD matches the dLLM's denoiser logits to the teacher's via a normalization-free objective, which reduces RL to likelihood-free self-distillation and thus bypasses the TIM biases. Recent ELBO-based methods emerge as instances of applying different distillation divergences, but with diagnosable pathologies that GDSD avoids. On planning, math, and coding benchmarks with LLaDA-8B and Dream-7B, GDSD consistently outperforms prior state-of-the-art ELBO-based methods with a more stable training reward dynamics, achieving test-accuracy improvements of up to $+19.6\%$. These results suggest that direct denoiser self-distillation, without relying on an ELBO likelihood surrogate, can provide a more stable and effective RL procedure for dLLMs. Code is available at https://github.com/GaryBall/GDSD.

Kaipeng Xu, Zhuo Zhi, Keyue Jiang

Coordinating large-scale, heterogeneous building aggregations for demand response (DR) is impeded by a dual challenge: the computational intractability of centralized Model Predictive Control (MPC) and the inadequacy of conventional feature selection methods, which fail to address the error-compounding nature of multi-step forecasting required by MPC. This paper proposes a comprehensive, data-driven framework that first employs a systematic, MPC-aware feature selection methodology to ensure robust multi-step prediction, then models the complex building dynamics using a novel Input-Convex Encoder-Only Transformer (IC-EoT) to guarantee a convex optimization problem, and finally solves the resulting constraint-coupled problem (CCP) in a fully distributed manner using the Tracking Alternating Direction Method of Multipliers (ADMM) algorithm. The framework is validated in a high-fidelity co-simulation environment, controlling a heterogeneous aggregation of consumer and prosumer buildings based on the EnergyPlus under a dynamic time-of-use (TOU) tariff. Results demonstrate that the proposed distributed approach achieves near-identical economic optimality and superior thermal comfort compared to a theoretical centralized controller, while exhibiting exceptional computational scalability that overcomes the real-time infeasibility of the centralized approach for large aggregations.

Tommaso Giorgi, Pierriccardo Olivieri, Keyue Jiang et al.

Learning compact state representations in Markov Decision Processes (MDPs) has proven crucial for addressing the curse of dimensionality in large-scale reinforcement learning (RL) problems. Existing principled approaches leverage structural priors on the MDP by constructing state representations as linear combinations of the state-graph Laplacian eigenvectors. When the transition graph is unknown or the state space is prohibitively large, the graph spectral features can be estimated directly via sample trajectories. In this work, we prove an upper bound on the approximation error of linear value function approximation under the learned spectral features. We show how this error scales with the algebraic connectivity of the state-graph, grounding the approximation quality in the topological structure of the MDP. We further bound the error introduced by the eigenvector estimation itself, leading to an end-to-end error decomposition across the representation learning pipeline. Additionally, our expression of the Laplacian operator for the RL setting, although equivalent to existing ones, prevents some common misunderstandings, of which we show some examples from the literature. Our results hold for general (non-uniform) policies without any assumptions on the symmetry of the induced transition kernel. We validate our theoretical findings with numerical simulations on gridworld environments.

Bohan Tang, Keyue Jiang, Laura Toni et al.

Understanding the data-generating process is essential for building machine learning models that generalise well while ensuring robustness and interpretability. This paper addresses the fundamental challenge of modelling the data generation processes on hypergraphs and explores how such models can inform the design of machine learning algorithms for hypergraph data. The key to our approach is the development of a hypergraph Markov random field that models the joint distribution of the node features and hyperedge features in a hypergraph through a multivariate Gaussian distribution whose covariance matrix is uniquely determined by the hypergraph structure. The proposed data-generating process provides a valuable inductive bias for various hypergraph machine learning tasks, thus enhancing the algorithm design. In this paper, we focus on two representative downstream tasks: structure inference and node classification. Accordingly, we introduce two novel frameworks: 1) an original hypergraph structure inference framework named HGSI, and 2) a novel learning framework entitled Hypergraph-MLP for node classification on hypergraphs. Empirical evaluation of the proposed frameworks demonstrates that: 1) HGSI outperforms existing hypergraph structure inference methods on both synthetic and real-world data; and 2) Hypergraph-MLP outperforms baselines in six hypergraph node classification benchmarks, at the same time promoting runtime efficiency and robustness against structural perturbations during inference.

Tiantian He, Yihang Chen, Keyue Jiang et al.

Computer-use agents that combine GUI interaction with structured API calls via the Model Context Protocol (MCP) show promise for automating software tasks. However, existing approaches lack a principled understanding of how agents should balance these two modalities and how to enable iterative self-improvement across diverse applications. We formulate MCP-GUI interplay as a unified hybrid policy learning problem where the agent learns when each modality provides complementary advantages, and show that distillation and experience augmentation target fundamentally different failure modes - requiring application-aware mechanism selection. Built on this formulation, we propose a self-evolving framework with a fully automatic pipeline that orchestrates automatic environment generation and validation, trajectory collection, gap-driven task synthesis, and quality-filtered training - all without manual intervention. A key innovation is our experience bank, which accumulates LLM-learned rules from trajectory comparison, enabling inference-time improvement without fine-tuning. Systematic \textbf{cross-application analysis} across three desktop applications reveals that the optimal strategy depends on MCP-GUI composition: distillation achieves 77.8\% pass rate on MCP-dominant tasks (+17.8pp), while the experience bank excels on GUI-intensive tasks (+10.0pp).

Yanan Zhao, Feng Ji, Jingyang Dai et al.

Graph contrastive learning (GCL) learns node and graph representations by contrasting multiple views of the same graph. Existing methods typically rely on fixed, handcrafted views-usually a local and a global perspective, which limits their ability to capture multi-scale structural patterns. We present an augmentation-free, multi-view GCL framework grounded in fractional-order continuous dynamics. By varying the fractional derivative order $α\in (0,1]$, our encoders produce a continuous spectrum of views: small $α$ yields localized features, while large $α$ induces broader, global aggregation. We treat $α$ as a learnable parameter so the model can adapt diffusion scales to the data and automatically discover informative views. This principled approach generates diverse, complementary representations without manual augmentations. Extensive experiments on standard benchmarks demonstrate that our method produces more robust and expressive embeddings and outperforms state-of-the-art GCL baselines.

Yuxiang Wang, Yu Xiang, Baojian Zhou et al.

Masked diffusion language models (MDMs) have recently emerged as a promising alternative to standard autoregressive large language models (AR-LLMs), yet their optimization can be substantially less stable. We study blockwise MDMs and compare them with AR-LLMs on three controlled tasks that stress different aspects of structured generation: in-context linear regression, graph path-finding, and Sudoku solving. We find that standard random-masking MDMs fail to reliably learn linear regression, exhibit high variance training dynamics on graph path-finding, while outperforming AR-LLMs on Sudoku. To mitigate these instabilities, we propose two locality aware blockwise models, namely Jigsaw and Scatter, that inject left-to-right inductive bias by enforcing autoregressive locality within blocks while preserving iterative refinement at the block level. Empirically, Jigsaw matches AR-LLM stability on linear regression and remains strong on Sudoku, while Scatter retains diffusion's planning advantage on path-finding. Our results indicate that standard random-masking MDMs, even with blockwise variants, may be a suboptimal instantiation of diffusion LMs for ordered generation, motivating models beyond random masking.

Mingyu Xu, Cheng Fang, Keyue Jiang et al.

We present Logics-STEM, a state-of-the-art reasoning model fine-tuned on Logics-STEM-SFT-Dataset, a high-quality and diverse dataset at 10M scale that represents one of the largest-scale open-source long chain-of-thought corpora. Logics-STEM targets reasoning tasks in the domains of Science, Technology, Engineering, and Mathematics (STEM), and exhibits exceptional performance on STEM-related benchmarks with an average improvement of 4.68% over the next-best model at 8B scale. We attribute the gains to our data-algorithm co-design engine, where they are jointly optimized to fit a gold-standard distribution behind reasoning. Data-wise, the Logics-STEM-SFT-Dataset is constructed from a meticulously designed data curation engine with 5 stages to ensure the quality, diversity, and scalability, including annotation, deduplication, decontamination, distillation, and stratified sampling. Algorithm-wise, our failure-driven post-training framework leverages targeted knowledge retrieval and data synthesis around model failure regions in the Supervised Fine-tuning (SFT) stage to effectively guide the second-stage SFT or the reinforcement learning (RL) for better fitting the target distribution. The superior empirical performance of Logics-STEM reveals the vast potential of combining large-scale open-source data with carefully designed synthetic data, underscoring the critical role of data-algorithm co-design in enhancing reasoning capabilities through post-training. We make both the Logics-STEM models (8B and 32B) and the Logics-STEM-SFT-Dataset (10M and downsampled 2.2M versions) publicly available to support future research in the open-source community.

Nagham Osman, Keyue Jiang, Davide Buffelli et al.

Graph generation is a critical task across scientific domains. Existing methods fall broadly into two categories: autoregressive models, which iteratively expand graphs, and one-shot models, such as diffusion, which generate the full graph at once. In this work, we provide an analysis of these two paradigms and reveal a key trade-off: autoregressive models stand out in capturing fine-grained local structures, such as degree and clustering properties, whereas one-shot models excel at modeling global patterns, such as spectral distributions. Building on this, we propose LGDC (latent graph diffusion via spectrum-preserving coarsening), a hybrid framework that combines strengths of both approaches. LGDC employs a spectrum-preserving coarsening-decoarsening to bidirectionally map between graphs and a latent space, where diffusion efficiently generates latent graphs before expansion restores detail. This design captures both local and global properties with improved efficiency. Empirically, LGDC matches autoregressive models on locally structured datasets (Tree) and diffusion models on globally structured ones (Planar, Community-20), validating the benefits of hybrid generation.

Kaipeng Xu, Zhuo Zhi, Keyue Jiang

Learning-based Model Predictive Control (MPC) has emerged as a powerful strategy for building demand response. However, its practical deployment is often hindered by the non-convex optimization problems induced by standard neural network models. These problems lead to long solver times and suboptimal solutions, making real-time control over long horizons challenging. While Input Convex Neural Networks (ICNNs), such as Input-Convex Long Short-Term Memorys (IC-LSTMs), are developed to address the convexity issue, their recurrent architectures suffer from high computational cost and gradient instability as the prediction horizon increases. To overcome these limitations, this paper introduces the Input-Convex Encoder-only Transformer (IC-EoT), a novel architecture that synergizes the parallel processing capabilities of the Transformer with the guaranteed tractability of input convexity. The IC-EoT was developed and evaluated in a high-fidelity co-simulation framework using the Energym Python library to interface with the EnergyPlus building simulator, and compared against its recurrent convex counterpart (IC-LSTM) and standard non-convex models. The results demonstrate that the IC-EoT is structurally immune to the gradient instability that affects recurrent ICNNs while maintaining comparable predictive accuracy. More critically, it substantially reduces MPC solver times; this speed advantage grows with the prediction horizon, with the IC-EoT proving 2.7 to 8.3 times faster than the IC-LSTM across horizons spanning from one to eight hours. This leap in computational efficiency makes the IC-EoT a robust and practical solution, enabling effective, real-time MPC for building energy management under realistic horizon decision-making scenarios.

Kaipeng Xu, Zhuo Zhi, Ruixuan Zhao et al.

The high energy consumption of buildings presents a critical need for advanced control strategies like Demand Response (DR). Differentiable Predictive Control (DPC) has emerged as a promising method for learning explicit control policies, yet conventional DPC frameworks are hindered by three key limitations: the use of simplistic dynamics models with limited expressiveness, a decoupled training paradigm that fails to optimize for closed-loop performance, and a lack of practical safety guarantees under realistic assumptions. To address these shortcomings, this paper proposes a novel End-to-End Differentiable Predictive Control (E2E-DPC) framework. Our approach utilizes an Encoder-Only Transformer to model the complex system dynamics and employs a unified, performance-oriented loss to jointly train the model and the control policy. Crucially, we introduce an online tube-based constraint tightening method that provides theoretical guarantees for recursive feasibility and constraint satisfaction without requiring complex offline computation of terminal sets. The framework is validated in a high-fidelity EnergyPlus simulation, controlling a multi-zone building for a DR task. The results demonstrate that the proposed method with guarantees achieves near-perfect constraint satisfaction - a reduction of over 99% in violations compared to the baseline - at the cost of only a minor increase in electricity expenditure. This work provides a deployable, performance-driven control solution for building energy management and establishes a new pathway for developing verifiable learning-based control systems under milder assumptions.

Bohan Tang, Zexi Liu, Keyue Jiang et al.

Hypergraphs are crucial for modelling higher-order interactions in real-world data. Hypergraph neural networks (HNNs) effectively utilise these structures by message passing to generate informative node features for various downstream tasks like node classification. However, the message passing module in existing HNNs typically requires a computationally intensive training process, which limits their practical use. To tackle this challenge, we propose an alternative approach by decoupling the usage of hypergraph structural information from the model learning stage. This leads to a novel training-free message passing module, named TF-MP-Module, which can be precomputed in the data preprocessing stage, thereby reducing the computational burden. We refer to the hypergraph neural network equipped with our TF-MP-Module as TF-HNN. We theoretically support the efficiency and effectiveness of TF-HNN by showing that: 1) It is more training-efficient compared to existing HNNs; 2) It utilises as much information as existing HNNs for node feature generation; and 3) It is robust against the oversmoothing issue while using long-range interactions. Experiments based on seven real-world hypergraph benchmarks in node classification and hyperlink prediction show that, compared to state-of-the-art HNNs, TF-HNN exhibits both competitive performance and superior training efficiency. Specifically, on the large-scale benchmark, Trivago, TF-HNN outperforms the node classification accuracy of the best baseline by 10% with just 1% of the training time of that baseline.

Keyue Jiang, Bohan Tang, Xiaowen Dong et al.

Inferring the graph structure from observed data is a key task in graph machine learning to capture the intrinsic relationship between data entities. While significant advancements have been made in learning the structure of homogeneous graphs, many real-world graphs exhibit heterogeneous patterns where nodes and edges have multiple types. This paper fills this gap by introducing the first approach for heterogeneous graph structure learning (HGSL). To this end, we first propose a novel statistical model for the data-generating process (DGP) of heterogeneous graph data, namely hidden Markov networks for heterogeneous graphs (H2MN). Then we formalize HGSL as a maximum a-posterior estimation problem parameterized by such DGP and derive an alternating optimization method to obtain a solution together with a theoretical justification of the optimization conditions. Finally, we conduct extensive experiments on both synthetic and real-world datasets to demonstrate that our proposed method excels in learning structure on heterogeneous graphs in terms of edge type identification and edge weight recovery.

Keyue Jiang, Jiahao Cui, Xiaowen Dong et al.

Graph generation has emerged as a critical task in fields ranging from drug discovery to circuit design. Contemporary approaches, notably diffusion and flow-based models, have achieved solid graph generative performance through constructing a probability path that interpolates between reference and data distributions. However, these methods typically model the evolution of individual nodes and edges independently and use linear interpolations to build the path. This disentangled interpolation breaks the interconnected patterns of graphs, making the constructed probability path irregular and non-smooth, which causes poor training dynamics and faulty sampling convergence. To address the limitation, this paper first presents a theoretically grounded framework for probability path construction in graph generative models. Specifically, we model the joint evolution of the nodes and edges by representing graphs as connected systems parameterized by Markov random fields (MRF). We then leverage the optimal transport displacement between MRF objects to design a smooth probability path that ensures the co-evolution of graph components. Based on this, we introduce BWFlow, a flow-matching framework for graph generation that utilizes the derived optimal probability path to benefit the training and sampling algorithm design. Experimental evaluations in plain graph generation and molecule generation validate the effectiveness of BWFlow with competitive performance, better training convergence, and efficient sampling.

Jasraj Singh, Keyue Jiang, Brooks Paige et al.

Message Passing Neural Networks (MPNNs) are a class of Graph Neural Networks (GNNs) that propagate information across the graph via local neighborhoods. The scheme gives rise to two key challenges: over-smoothing and over-squashing. While several Dropout-style algorithms, such as DropEdge and DropMessage, have successfully addressed over-smoothing, their impact on over-squashing remains largely unexplored. This represents a critical gap in the literature, as failure to mitigate over-squashing would make these methods unsuitable for long-range tasks -- the intended use case of deep MPNNs. In this work, we study the aforementioned algorithms, and closely related edge-dropping algorithms -- DropNode, DropAgg and DropGNN -- in the context of over-squashing. We present theoretical results showing that DropEdge-variants reduce sensitivity between distant nodes, limiting their suitability for long-range tasks. To address this, we introduce DropSens, a sensitivity-aware variant of DropEdge that explicitly controls the proportion of information lost due to edge-dropping, thereby increasing sensitivity to distant nodes despite dropping the same number of edges. Our experiments on long-range synthetic and real-world datasets confirm the predicted limitations of existing edge-dropping and feature-dropping methods. Moreover, DropSens consistently outperforms graph rewiring techniques designed to mitigate over-squashing, suggesting that simple, targeted modifications can substantially improve a model's ability to capture long-range interactions. Our conclusions highlight the need to re-evaluate and re-design existing methods for training deep GNNs, with a renewed focus on modelling long-range interactions.

Bohan Tang, Huidong Liang, Keyue Jiang et al.

Large language model multi-agent systems (LLM-MAS) offer a promising paradigm for harnessing collective intelligence to achieve more advanced forms of AI behaviour. While recent studies suggest that LLM-MAS can outperform LLM single-agent systems (LLM-SAS) on certain tasks, the lack of systematic experimental designs limits the strength and generality of these conclusions. We argue that a principled understanding of task complexity, such as the degree of sequential reasoning required and the breadth of capabilities involved, is essential for assessing the effectiveness of LLM-MAS in task solving. To this end, we propose a theoretical framework characterising tasks along two dimensions: depth, representing reasoning length, and width, representing capability diversity. We theoretically examine a representative class of LLM-MAS, namely the multi-agent debate system, and empirically evaluate its performance in both discriminative and generative tasks with varying depth and width. Theoretical and empirical results show that the benefit of LLM-MAS over LLM-SAS increases with both task depth and width, and the effect is more pronounced with respect to depth. This clarifies when LLM-MAS are beneficial and provides a principled foundation for designing future LLM-MAS methods and benchmarks.

Tiantian He, Keyue Jiang, An Zhao et al.

The long-term progression of neurodegenerative diseases is commonly conceptualized as a spatiotemporal diffusion process that consists of a graph diffusion process across the structural brain connectome and a localized reaction process within brain regions. However, modeling this progression remains challenging due to 1) the scarcity of longitudinal data obtained through irregular and infrequent subject visits and 2) the complex interplay of pathological mechanisms across brain regions and disease stages, where traditional models assume fixed mechanisms throughout disease progression. To address these limitations, we propose a novel stage-aware Mixture of Experts (MoE) framework that explicitly models how different contributing mechanisms dominate at different disease stages through time-dependent expert weighting.Data-wise, we utilize an iterative dual optimization method to properly estimate the temporal position of individual observations, constructing a co hort-level progression trajectory from irregular snapshots. Model-wise, we enhance the spatial component with an inhomogeneous graph neural diffusion model (IGND) that allows diffusivity to vary based on node states and time, providing more flexible representations of brain networks. We also introduce a localized neural reaction module to capture complex dynamics beyond standard processes.The resulting IGND-MoE model dynamically integrates these components across temporal states, offering a principled way to understand how stage-specific pathological mechanisms contribute to progression. The stage-wise weights yield novel clinical insights that align with literature, suggesting that graph-related processes are more influential at early stages, while other unknown physical processes become dominant later on.