arXiv subjects
Publications and source records attributed to Yufeng Jiang.
We present the first LLM-powered multi-agent system for autonomous fault management across millions of optical links in production AIDCs. Refined via SFT and continuous memory evolution, it achieves 97.7% F1 and over 60% fault-incident reduction, outperforming SOTA LLMs on a ten-week field data evaluation.
Automatic speech recognition (ASR) in clinical dialogue demands robustness to full-duplex interaction, speaker overlap, and low-latency constraints, yet open benchmarks remain scarce. We present MMedFD, the first real-world Chinese healthcare ASR corpus designed for multi-turn, full-duplex settings. Captured from a deployed AI assistant, the dataset comprises 5,805 annotated sessions with synchronized user and mixed-channel views, RTTM/CTM timing, and role labels. We introduce a model-agnostic pipeline for streaming segmentation, speaker attribution, and dialogue memory, and fine-tune Whisper-small on role-concatenated audio for long-context recognition. ASR evaluation includes WER, CER, and HC-WER, which measures concept-level accuracy across healthcare settings. LLM-generated responses are assessed using rubric-based and pairwise protocols. MMedFD establishes a reproducible framework for benchmarking streaming ASR and end-to-end duplex agents in healthcare deployment. The dataset and related resources are publicly available at https://github.com/Kinetics-JOJO/MMedFD
Accurate prediction of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) is important for drug discovery. Most predictors use undirected molecular graphs and pairwise edges. This choice misses asymmetric interactions, nonreversible dynamics, and motif level effects from functional groups and ring systems. We propose ChemHyperMag for multitask ADMET prediction under missing labels. ChemHyperMag builds a functional group hypergraph from rings, BRICS fragments, Bemis-Murcko scaffolds, and bonds. It also defines a potential driven nonreversible flow guided by electronegativity and Gasteiger partial charges. The resulting circulation is encoded by a Hermitian magnetic Laplacian and processed with a magnetic Chebyshev encoder. We perturb magnetic phases to form stochastic views and train with an InfoNCE objective. Experiments on multiple ADMET benchmarks show improvements over recent methods with fewer labeled samples and no conformers. ChemHyperMag is scalable and provides interpretable directional signals through its magnetic phases.
Alzheimer's disease (AD) manifests as a continuous progression from normal cognition (NC) through mild cognitive impairment (MCI) to dementia. However, most deep learning approaches reduce this continuum to disjointed classification tasks, largely ignoring dynamic stage transitions. To decode this complex progression, we propose M$^3$AD, a unified framework that jointly addresses three-class diagnosis classification and diagnosis stage transition prediction using only T1-weighted sMRI. M$^3$AD leverages an interpretable multi-gate mixture of experts architecture, employing specialized routing mechanisms to dynamically capture both diagnosis-specific pathological patterns and shared structural features across the continuum. It further integrates clinical priors (age, sex, eTIV) via adaptive attention fusion to enhance generalization. M$^3$AD achieves 95.13% accuracy, compared to 90.44% reported by MCLNC under its original experimental setting, and 94.87% for transition prediction. Crucially, analyzing the multi-gate routing reveals distinct expert activation signatures distinguishing stable from progressive MCI, providing a mechanistic basis for individual-level progression risk stratification. Code is available at https://github.com/csyfjiang/M3AD.
Accurate identification of drug-target interactions (DTI) remains a central challenge in computational pharmacology, where sequence-based methods offer scalability. This work introduces a sequence-based drug-target interaction framework that integrates structural priors into protein representations while maintaining high-throughput screening capability. Evaluated across multiple benchmarks, the model achieves state-of-the-art performance on Human and BioSNAP datasets and remains competitive on BindingDB. In virtual screening tasks, it surpasses prior methods on LIT-PCBA, yielding substantial gains in AUROC and BEDROC. Ablation studies confirm the critical role of learned aggregation, bilinear attention, and contrastive alignment in enhancing predictive robustness. Embedding visualizations reveal improved spatial correspondence with known binding pockets and highlight interpretable attention patterns over ligand-residue contacts. These results validate the framework's utility for scalable and structure-aware DTI prediction.
AI models for drug discovery and chemical literature mining must interpret molecular images and generate outputs consistent with 3D geometry and stereochemistry. Most molecular language models rely on strings or graphs, while vision-language models often miss stereochemical details and struggle to map continuous 3D structures into discrete tokens. We propose DeepMoLM: Deep Molecular Language M odeling, a dual-view framework that grounds high-resolution molecular images in geometric invariants derived from molecular conformations. DeepMoLM preserves high-frequency evidence from 1024 $\times$ 1024 inputs, encodes conformer neighborhoods as discrete Extended 3-Dimensional Fingerprints, and fuses visual and geometric streams with cross-attention, enabling physically grounded generation without atom coordinates. DeepMoLM improves PubChem captioning with a 12.3% relative METEOR gain over the strongest generalist baseline while staying competitive with specialist methods. It produces valid numeric outputs for all property queries and attains MAE 13.64 g/mol on Molecular Weight and 37.89 on Complexity in the specialist setting. On ChEBI-20 description generation from images, it exceeds generalist baselines and matches state-of-the-art vision-language models. Code is available at https://github.com/1anj/DeepMoLM.
Stream processing is a computing paradigm that supports real-time data processing for a wide variety of applications. At Meta, it's used across the company for various tasks such as deriving product insights, providing and improving user services, and enabling AI at scale for our ever-growing user base. Meta's current stream processing framework supports processing TerraBytes(TBs) of data in mere seconds. This is enabled by our efficient schedulers and multi-layered infrastructure, which allocate workloads across various compute resources, working together in hierarchies across various parts of the infrastructure. But with the ever growing complexity of applications, and user needs, areas of the infrastructure that previously required minimal load balancing, now must be made more robust and proactive to application load. In our work we explore how to build and design such a system that focuses on load balancing over key compute resources and properties of these applications. We also showcase how to integrate new schedulers into the hierarchy of the existing ones, allowing multiple schedulers to work together and perform load balancing, at their infrastructure level, effectively.
Reliable and interpretable tumor classification from clinical imaging remains a core challenge. The main difficulties arise from heterogeneous modality quality, limited annotations, and the absence of structured anatomical guidance. We present REACT-KD, a Region-Aware Cross-modal Topological Knowledge Distillation framework that transfers supervision from high-fidelity multi-modal sources into a lightweight CT-based student model. The framework employs a dual teacher design. One branch captures structure-function relationships through dual-tracer PET/CT, while the other models dose-aware features using synthetically degraded low-dose CT. These branches jointly guide the student model through two complementary objectives. The first achieves semantic alignment through logits distillation, and the second models anatomical topology through region graph distillation. A shared CBAM3D module ensures consistent attention across modalities. To improve reliability in deployment, REACT-KD introduces modality dropout during training, which enables robust inference under partial or noisy inputs. As a case study, we applied REACT-KD to hepatocellular carcinoma staging. The framework achieved an average AUC of 93.5\% on an internal PET/CT cohort and maintained 76.6\% to 81.5\% AUC across varying levels of dose degradation in external CT testing. Decision curve analysis further shows that REACT-KD consistently provides the highest net clinical benefit across all thresholds, confirming its value in real-world diagnostic practice. Code is available at: https://github.com/Kinetics-JOJO/REACT-KD
Accurate prediction of protein-ligand interactions is essential for computer-aided drug discovery. However, existing methods often fail to capture solvent-dependent conformational changes and lack the ability to jointly learn multiple related tasks. To address these limitations, we introduce a pre-training method that incorporates ligand conformational ensembles generated under diverse solvent conditions as augmented input. This design enables the model to learn both structural flexibility and environmental context in a unified manner. The training process integrates molecular reconstruction to capture local geometry, interatomic distance prediction to model spatial relationships, and contrastive learning to build solvent-invariant molecular representations. Together, these components lead to significant improvements, including a 3.7% gain in binding affinity prediction, an 82% success rate on the PoseBusters Astex docking benchmarks, and an area under the curve of 97.1% in virtual screening. The framework supports solvent-aware, multi-task modeling and produces consistent results across benchmarks. A case study further demonstrates sub-angstrom docking accuracy with a root-mean-square deviation of 0.157 angstroms, offering atomic-level insight into binding mechanisms and advancing structure-based drug design.
Recent advancements in medical imaging have resulted in more complex and diverse images, with challenges such as high anatomical variability, blurred tissue boundaries, low organ contrast, and noise. Traditional segmentation methods struggle to address these challenges, making deep learning approaches, particularly U-shaped architectures, increasingly prominent. However, the quadratic complexity of standard self-attention makes Transformers computationally prohibitive for high-resolution images. To address these challenges, we propose MLLA-UNet (Mamba-Like Linear Attention UNet), a novel architecture that achieves linear computational complexity while maintaining high segmentation accuracy through its innovative combination of linear attention and Mamba-inspired adaptive mechanisms, complemented by an efficient symmetric sampling structure for enhanced feature processing. Our architecture effectively preserves essential spatial features while capturing long-range dependencies at reduced computational complexity. Additionally, we introduce a novel sampling strategy for multi-scale feature fusion. Experiments demonstrate that MLLA-UNet achieves state-of-the-art performance on six challenging datasets with 24 different segmentation tasks, including but not limited to FLARE22, AMOS CT, and ACDC, with an average DSC of 88.32%. These results underscore the superiority of MLLA-UNet over existing methods. Our contributions include the novel 2D segmentation architecture and its empirical validation. The code is available via https://github.com/csyfjiang/MLLA-UNet.
Medical imaging data is inherently heterogeneous across different modalities and clinical centers, posing unique challenges for developing generalizable foundation models. Conventional entails training distinct models per dataset or using a shared encoder with modality-specific decoders. However, these approaches incur heavy computational overheads and suffer from poor scalability. To address these limitations, we propose the Medical Multimodal Mixture of Experts (M$^4$oE) framework, leveraging the SwinUNet architecture. Specifically, M$^4$oE comprises modality-specific experts; each separately initialized to learn features encoding domain knowledge. Subsequently, a gating network is integrated during fine-tuning to modulate each expert's contribution to the collective predictions dynamically. This enhances model interpretability and generalization ability while retaining expertise specialization. Simultaneously, the M$^4$oE architecture amplifies the model's parallel processing capabilities, and it also ensures the model's adaptation to new modalities with ease. Experiments across three modalities reveal that M$^4$oE can achieve 3.45% over STU-Net-L, 5.11% over MED3D, and 11.93% over SAM-Med2D across the MICCAI FLARE22, AMOS2022, and ATLAS2023 datasets. Moreover, M$^4$oE showcases a significant reduction in training duration with 7 hours less while maintaining a parameter count that is only 30% of its compared methods. The code is available at https://github.com/JefferyJiang-YF/M4oE.
Red and blue particles are placed in equal proportion through-out either the complete or star graph and iteratively sampled to take simple random walk steps. Mutual annihilation occurs when particles with different colors meet. We compare the time it takes to extinguish every particle to the analogous time in the (simple to analyze) one-type setting. Additionally, we study the effect of asymmetric particle speeds.
We consider evoSIR, a variant of the SIR model, on Erd\H os-Renyi random graphs in which susceptibles with an infected neighbor break that connection at rate $ρ$ and rewire to a randomly chosen individual. We compute the critical infection rate $λ_c$ and the probability of a large epidemic by showing that they are the same for the delSIR model in which $S-I$ connections are deleted instead of rewired. The final size of a large delSIR epidemic has a continuous transition. Simulations suggest that the final size of a large evoSIR epidemic is discontinuous at $λ_c$.
A little over 25 years ago Pemantle pioneered the study of the contact process on trees, and showed that the critical values $λ_1$ and $λ_2$ for global and local survival were different. Here, we will consider the case of trees in which the degrees of vertices are periodic. We will compute bounds on $λ_1$ and $λ_2$ and for the corresponding critical values $λ_g$ and $λ_\ell$ for branching random walk. Much of what we find for period two $(a,b)$ trees was known to Pemantle. However, two significant new results give sharp asymptotics for the critical value $λ_2$ of $(1,n)$ trees and generalize that result to the $(a_1,\ldots, a_k, n)$ tree when $\max_i a_i \le n^{1-ε}$ and $a_1 \cdots a_k = n^b$. We also give results for $λ_g$ and $λ_\ell$ on $(a,b,c)$ trees. Since the values come from solving cubic equations, the explicit formulas are not pretty, but it is surprising that they depend only on $a+b+c$ and $abc$.
We provide a uniform upper bound on the minimal drift so that the one-per-site frog model on a $d$-ary tree is recurrent. To do this, we introduce a subprocess that couples across trees with different degrees. Finding couplings for frog models on nested sequences of graphs is known to be difficult. The upper bound comes from combining the coupling with a new, simpler proof that the frog model on a binary tree is recurrent when the drift is sufficiently strong. Additionally, we describe a coupling between frog models on trees for which the degree of the smaller tree divides that of the larger one. This implies that the critical drift has a limit as $d$ tends to infinity along certain subsequences.