Jiayi Huang

Heyang Liu, Ziyang Cheng, Jiayi Huang et al.

Recent advances in Speech Large Language Models (Speech LLMs) have significantly enhanced spoken language understanding and reasoning. However, their contextual awareness is limited, struggling to perform speech recognition that effectively reflects the speaker's intent and topical context. In this paper, we propose LaSR (Latent Speech Reasoning), a novel training paradigm featuring a context-aware reasoning trajectory that leverages the latent reasoning process. Instead of generating explicit intermediate tokens, LaSR aligns chain-of-thought (CoT) supervision around the acoustic feature region of the targeted word, and introduces latent reasoning periods for context information grounding and transcriptional transition. Furthermore, to effectively benchmark contextual recognition on specialized vocabulary, we propose Spoken Darwin-Science, a large-scale corpus focusing on academic terminologies. Preliminary experiments on Fun-Audio-Chat demonstrate that LaSR significantly improves terminology recognition without introducing additional latency and consistently outperforms standard supervised fine-tuning baselines. Our findings highlight the potential of latent reasoning in building efficient, context-aware speech assistants.

Jiayi Huang, Han Zhong, Liwei Wang et al.

While numerous works have focused on devising efficient algorithms for reinforcement learning (RL) with uniformly bounded rewards, it remains an open question whether sample or time-efficient algorithms for RL with large state-action space exist when the rewards are \emph{heavy-tailed}, i.e., with only finite $(1+ε)$-th moments for some $ε\in(0,1]$. In this work, we address the challenge of such rewards in RL with linear function approximation. We first design an algorithm, \textsc{Heavy-OFUL}, for heavy-tailed linear bandits, achieving an \emph{instance-dependent} $T$-round regret of $\tilde{O}\big(d T^{\frac{1-ε}{2(1+ε)}} \sqrt{\sum_{t=1}^T ν_t^2} + d T^{\frac{1-ε}{2(1+ε)}}\big)$, the \emph{first} of this kind. Here, $d$ is the feature dimension, and $ν_t^{1+ε}$ is the $(1+ε)$-th central moment of the reward at the $t$-th round. We further show the above bound is minimax optimal when applied to the worst-case instances in stochastic and deterministic linear bandits. We then extend this algorithm to the RL settings with linear function approximation. Our algorithm, termed as \textsc{Heavy-LSVI-UCB}, achieves the \emph{first} computationally efficient \emph{instance-dependent} $K$-episode regret of $\tilde{O}(d \sqrt{H \mathcal{U}^*} K^\frac{1}{1+ε} + d \sqrt{H \mathcal{V}^* K})$. Here, $H$ is length of the episode, and $\mathcal{U}^*, \mathcal{V}^*$ are instance-dependent quantities scaling with the central moment of reward and value functions, respectively. We also provide a matching minimax lower bound $Ω(d H K^{\frac{1}{1+ε}} + d \sqrt{H^3 K})$ to demonstrate the optimality of our algorithm in the worst case. Our result is achieved via a novel robust self-normalized concentration inequality that may be of independent interest in handling heavy-tailed noise in general online regression problems.

Weilin Cai, Le Qin, Jiayi Huang

As large language models continue to scale up, distributed training systems have expanded beyond 10k nodes, intensifying the importance of fault tolerance. Checkpoint has emerged as the predominant fault tolerance strategy, with extensive studies dedicated to optimizing its efficiency. However, the advent of the sparse Mixture-of-Experts (MoE) model presents new challenges due to the substantial increase in model size, despite comparable computational demands to dense models. In this work, we propose the Mixture-of-Checkpoint System (MoC-System) to orchestrate the vast array of checkpoint shards produced in distributed training systems. MoC-System features a novel Partial Experts Checkpointing (PEC) mechanism, an algorithm-system co-design that strategically saves a selected subset of experts, effectively reducing the MoE checkpoint size to levels comparable with dense models. Incorporating hybrid parallel strategies, MoC-System involves fully sharded checkpointing strategies to evenly distribute the workload across distributed ranks. Furthermore, MoC-System introduces a two-level checkpointing management method that asynchronously handles in-memory snapshots and persistence processes. We build MoC-System upon the Megatron-DeepSpeed framework, achieving up to a 98.9% reduction in overhead for each checkpointing process compared to the original method, during MoE model training with ZeRO-2 data parallelism and expert parallelism. Additionally, extensive empirical analyses substantiate that our methods enhance efficiency while maintaining comparable model accuracy, even achieving an average accuracy increase of 1.08% on downstream tasks.

Jiayi Huang

Reliable inference requires that artificial intelligence (AI) models provide trustworthy uncertainty estimates, not merely accurate predictions. Recent advances in Bayesian learning have made significant progress toward this goal, and growing concerns about computational overhead have jointly shifted the design criterion from reliability alone to the co-design of reliability and efficiency, i.e., reducing computational overhead while preserving trustworthy uncertainty quantification. This thesis develops a unified framework from two perspectives to address the central question: can we efficiently perform reliable inference?

Weilin Cai, Juyong Jiang, Fan Wang et al.

Large language models (LLMs) have garnered unprecedented advancements across diverse fields, ranging from natural language processing to computer vision and beyond. The prowess of LLMs is underpinned by their substantial model size, extensive and diverse datasets, and the vast computational power harnessed during training, all of which contribute to the emergent abilities of LLMs (e.g., in-context learning) that are not present in small models. Within this context, the mixture of experts (MoE) has emerged as an effective method for substantially scaling up model capacity with minimal computation overhead, gaining significant attention from academia and industry. Despite its growing prevalence, there lacks a systematic and comprehensive review of the literature on MoE. This survey seeks to bridge that gap, serving as an essential resource for researchers delving into the intricacies of MoE. We first briefly introduce the structure of the MoE layer, followed by proposing a new taxonomy of MoE. Next, we overview the core designs for various MoE models including both algorithmic and systemic aspects, alongside collections of available open-source implementations, hyperparameter configurations and empirical evaluations. Furthermore, we delineate the multifaceted applications of MoE in practice, and outline some potential directions for future research. To facilitate ongoing updates and the sharing of cutting-edge advances in MoE research, we have established a resource repository at https://github.com/withinmiaov/A-Survey-on-Mixture-of-Experts-in-LLMs.

Weilin Cai, Juyong Jiang, Le Qin et al.

Expert parallelism has emerged as a key strategy for distributing the computational workload of sparsely-gated mixture-of-experts (MoE) models across multiple devices, enabling the processing of increasingly large-scale models. However, the All-to-All communication inherent to expert parallelism poses a significant bottleneck, limiting the efficiency of MoE models. Although existing optimization methods partially mitigate this issue, they remain constrained by the sequential dependency between communication and computation operations. To address this challenge, we propose ScMoE, a novel shortcut-connected MoE architecture integrated with an overlapping parallelization strategy. ScMoE decouples communication from its conventional sequential ordering, enabling up to 100% overlap with computation. Compared to the prevalent top-2 MoE baseline, ScMoE achieves speedups of 1.49 times in training and 1.82 times in inference. Moreover, our experiments and analyses indicate that ScMoE not only achieves comparable but in some instances surpasses the model quality of existing approaches.

Jiayi Huang, Feiyun Zhao, Lieyang Chen

This study proposes an advanced method for surface defect detection in printed circuit boards (PCBs) using an improved YOLOv11 model enhanced with a generative adversarial network (GAN). The approach focuses on identifying six common defect types: missing hole, rat bite, open circuit, short circuit, burr, and virtual welding. By employing GAN to generate synthetic defect images, the dataset is augmented with diverse and realistic patterns, improving the model's ability to generalize, particularly for complex and infrequent defects like burrs. The enhanced YOLOv11 model is evaluated on a PCB defect dataset, demonstrating significant improvements in accuracy, recall, and robustness, especially when dealing with defects in complex environments or small targets. This research contributes to the broader field of electronic design automation (EDA), where efficient defect detection is a crucial step in ensuring high-quality PCB manufacturing. By integrating advanced deep learning techniques, this approach enhances the automation and precision of defect detection, reducing reliance on manual inspection and accelerating design-to-production workflows. The findings underscore the importance of incorporating GAN-based data augmentation and optimized detection architectures in EDA processes, providing valuable insights for improving reliability and efficiency in PCB defect detection within industrial applications.

Yingping Wang, Yi Wu, Xiangyu Wu et al.

Mixture-of-Experts (MoE) architectures are widely used in modern large language models and multimodal models. However, inference efficiency is often limited by highly dynamic and skewed expert workloads across different modalities. During the prefill stage with large batch sizes, vision tokens frequently dominate the input sequences. Under expert parallelism (EP), this leads to severe load imbalance, where a subset of devices becomes overloaded, reducing overall system throughput. We propose ReaLB, a real-time load balancing method for multimodal MoE (MMoE) inference that introduces zero scheduling overhead. ReaLB dynamically adjusts the computation precision of MoE experts at runtime on a per-EP-rank basis. For ranks dominated by vision-heavy experts, ReaLB assigns lower-precision computation to improve execution efficiency by exploiting FP4 Tensor Cores. ReaLB does not require redundant experts or additional memory allocation. Instead, it performs layer-wise expert precision transformation on the fly and hides the associated overhead within the dispatch phase before MoE computation. Experiments on representative MMoE models show that ReaLB achieves 1.29x layer-level speedup while limiting accuracy loss to within 1.2%.

Jiayi Huang, Sangwoo Park, Osvaldo Simeone

The application of artificial intelligence (AI) models in fields such as engineering is limited by the known difficulty of quantifying the reliability of an AI's decision. A well-calibrated AI model must correctly report its accuracy on in-distribution (ID) inputs, while also enabling the detection of out-of-distribution (OOD) inputs. A conventional approach to improve calibration is the application of Bayesian ensembling. However, owing to computational limitations and model misspecification, practical ensembling strategies do not necessarily enhance calibration. This paper proposes an extension of variational inference (VI)-based Bayesian learning that integrates calibration regularization for improved ID performance, confidence minimization for OOD detection, and selective calibration to ensure a synergistic use of calibration regularization and confidence minimization. The scheme is constructed successively by first introducing calibration-regularized Bayesian learning (CBNN), then incorporating out-of-distribution confidence minimization (OCM) to yield CBNN-OCM, and finally integrating also selective calibration to produce selective CBNN-OCM (SCBNN-OCM). Selective calibration rejects inputs for which the calibration performance is expected to be insufficient. Numerical results illustrate the trade-offs between ID accuracy, ID calibration, and OOD calibration attained by both frequentist and Bayesian learning methods. Among the main conclusions, SCBNN-OCM is seen to achieve best ID and OOD performance as compared to existing state-of-the-art approaches at the cost of rejecting a sufficiently large number of inputs.

Shwai He, Weilin Cai, Jiayi Huang et al.

The Mixture of Experts (MoE) is an effective architecture for scaling large language models by leveraging sparse expert activation to balance performance and efficiency. However, under expert parallelism, MoE suffers from inference inefficiencies due to imbalanced token-to-expert assignment, where underloaded experts complete computations early but must wait for overloaded experts, leading to global delays. We define this phenomenon as the \textbf{\textit{Straggler Effect}}, as the most burdened experts dictate the overall inference latency. To address this, we first propose \textit{\textbf{Capacity-Aware Token Drop}}, which enforces expert capacity limits by discarding excess tokens from overloaded experts, effectively reducing load imbalance with minimal performance impact (e.g., $30\%$ speedup with only $0.9\%$ degradation on OLMoE). Next, given the presence of low-load experts remaining well below the capacity threshold, we introduce \textit{\textbf{Capacity-Aware Expanded Drop}}, which allows tokens to include additional local experts in their candidate set before enforcing strict local capacity constraints, thereby improving load balance and enhancing the utilization of underused experts. Extensive experiments on both language and multimodal MoE models demonstrate the effectiveness of our approach, yielding substantial gains in expert utilization, model performance, and inference efficiency, e.g., applying Expanded Drop to Mixtral-8$\times$7B-Instruct yields a {0.2\%} average performance improvement and a {1.85$\times$} inference speedup.

Jiayi Huang, Han Zhong, Liwei Wang et al.

To tackle long planning horizon problems in reinforcement learning with general function approximation, we propose the first algorithm, termed as UCRL-WVTR, that achieves both \emph{horizon-free} and \emph{instance-dependent}, since it eliminates the polynomial dependency on the planning horizon. The derived regret bound is deemed \emph{sharp}, as it matches the minimax lower bound when specialized to linear mixture MDPs up to logarithmic factors. Furthermore, UCRL-WVTR is \emph{computationally efficient} with access to a regression oracle. The achievement of such a horizon-free, instance-dependent, and sharp regret bound hinges upon (i) novel algorithm designs: weighted value-targeted regression and a high-order moment estimator in the context of general function approximation; and (ii) fine-grained analyses: a novel concentration bound of weighted non-linear least squares and a refined analysis which leads to the tight instance-dependent bound. We also conduct comprehensive experiments to corroborate our theoretical findings.

Shenghao Ren, Yi Lu, Jiayi Huang et al.

Existing human Motion Capture (MoCap) methods mostly focus on the visual similarity while neglecting the physical plausibility. As a result, downstream tasks such as driving virtual human in 3D scene or humanoid robots in real world suffer from issues such as timing drift and jitter, spatial problems like sliding and penetration, and poor global trajectory accuracy. In this paper, we revisit human MoCap from the perspective of interaction between human body and physical world by exploring the role of pressure. Firstly, we construct a large-scale human Motion capture dataset with Pressure, RGB and Optical sensors (named MotionPRO), which comprises 70 volunteers performing 400 types of motion, encompassing a total of 12.4M pose frames. Secondly, we examine both the necessity and effectiveness of the pressure signal through two challenging tasks: (1) pose and trajectory estimation based solely on pressure: We propose a network that incorporates a small kernel decoder and a long-short-term attention module, and proof that pressure could provide accurate global trajectory and plausible lower body pose. (2) pose and trajectory estimation by fusing pressure and RGB: We impose constraints on orthographic similarity along the camera axis and whole-body contact along the vertical axis to enhance the cross-attention strategy to fuse pressure and RGB feature maps. Experiments demonstrate that fusing pressure with RGB features not only significantly improves performance in terms of objective metrics, but also plausibly drives virtual humans (SMPL) in 3D scene. Furthermore, we demonstrate that incorporating physical perception enables humanoid robots to perform more precise and stable actions, which is highly beneficial for the development of embodied artificial intelligence. Project page is available at: https://nju-cite-mocaphumanoid.github.io/MotionPRO/

Jiayi Huang, Sangwoo Park, Nicola Paoletti et al.

Deploying artificial intelligence (AI) models on edge devices involves a delicate balance between meeting stringent complexity constraints, such as limited memory and energy resources, and ensuring reliable performance in sensitive decision-making tasks. One way to enhance reliability is through uncertainty quantification via Bayesian inference. This approach, however, typically necessitates maintaining and running multiple models in an ensemble, which may exceed the computational limits of edge devices. This paper introduces a low-complexity methodology to address this challenge by distilling calibration information from a more complex model. In an offline phase, predictive probabilities generated by a high-complexity cloud-based model are leveraged to determine a threshold based on the typical divergence between the cloud and edge models. At run time, this threshold is used to construct credal sets -- ranges of predictive probabilities that are guaranteed, with a user-selected confidence level, to include the predictions of the cloud model. The credal sets are obtained through thresholding of a divergence measure in the simplex of predictive probabilities. Experiments on visual and language tasks demonstrate that the proposed approach, termed Conformalized Distillation for Credal Inference (CD-CI), significantly improves calibration performance compared to low-complexity Bayesian methods, such as Laplace approximation, making it a practical and efficient solution for edge AI deployments.

Jiayi Huang, Sangwoo Park, Nicola Paoletti et al.

Edge intelligence enables low-latency inference via compact on-device models, but assuring reliability remains challenging. We study edge-cloud cascades that must preserve conditional coverage: whenever the edge returns a prediction set, it should contain the true label with a user-specified probability, as if produced by the cloud model. We formalize conditional coverage with respect to the cloud predictive distribution, and introduce a conformal alignment-based (CAb) cascading mechanism that certifies this property with user control over the risk level. Our method casts escalation from edge to cloud models as a multiple-hypothesis testing (MHT) problem, tailoring conformal alignment (CA) to select which inputs can be safely handled at the edge. The proposed CAb model cascading method yields statistical guarantees on the average fraction of edge decisions that satisfy cloud-level conditional coverage. The procedure applies to arbitrary edge prediction sets, including variants of conformal prediction (CP), and exposes a tunable trade-off among coverage, deferral rate, and set size. Experiments on CIFAR-100 image classification and the TeleQnA question-answering (QA) benchmark show that the proposed CAb cascade maintains the target conditional coverage for edge predictions while substantially reducing offloading to the cloud and incurring modest increases in prediction-set size.

Weilin Cai, Le Qin, Shwai He et al.

Mixture of Experts (MoE) has become a mainstream architecture for building Large Language Models (LLMs) by reducing per-token computation while enabling model scaling. It can be viewed as partitioning a large Feed-Forward Network (FFN) at the tensor level into fine-grained sub-FFNs, or experts, and activating only a sparse subset for each input. While this sparsity improves efficiency, MoE still faces substantial challenges due to their massive computational scale and unpredictable activation patterns. To enable efficient MoE deployment, we identify dual sparsity at the tensor and neuron levels in pre-trained MoE modules as a key factor for both accuracy and efficiency. Unlike prior work that increases tensor-level sparsity through finer-grained expert design during pre-training, we introduce post-training expert partitioning to induce such sparsity without retraining. This preserves the mathematical consistency of model transformations and enhances both efficiency and accuracy in subsequent fine-tuning and inference. Building upon this, we propose DualSparse-MoE, an inference system that integrates dynamic tensor-level computation dropping with static neuron-level reconstruction to deliver significant efficiency gains with minimal accuracy loss. Experimental results show that enforcing an approximate 25% drop rate with our approach reduces average accuracy by only 0.08%-0.28% across three prevailing MoE models, while nearly all degrees of computation dropping consistently yield proportional computational speedups. Furthermore, incorporating load-imbalance awareness into expert parallelism achieves a 1.41x MoE module speedup with just 0.5% average accuracy degradation.

Jiayi Huang, Sangwoo Park, Osvaldo Simeone

Deep learning models, including modern systems like large language models, are well known to offer unreliable estimates of the uncertainty of their decisions. In order to improve the quality of the confidence levels, also known as calibration, of a model, common approaches entail the addition of either data-dependent or data-independent regularization terms to the training loss. Data-dependent regularizers have been recently introduced in the context of conventional frequentist learning to penalize deviations between confidence and accuracy. In contrast, data-independent regularizers are at the core of Bayesian learning, enforcing adherence of the variational distribution in the model parameter space to a prior density. The former approach is unable to quantify epistemic uncertainty, while the latter is severely affected by model misspecification. In light of the limitations of both methods, this paper proposes an integrated framework, referred to as calibration-aware Bayesian neural networks (CA-BNNs), that applies both regularizers while optimizing over a variational distribution as in Bayesian learning. Numerical results validate the advantages of the proposed approach in terms of expected calibration error (ECE) and reliability diagrams.

Han Zhong, Jiayi Huang, Lin F. Yang et al.

Despite a large amount of effort in dealing with heavy-tailed error in machine learning, little is known when moments of the error can become non-existential: the random noise $η$ satisfies Pr$\left[|η| > |y|\right] \le 1/|y|^α$ for some $α> 0$. We make the first attempt to actively handle such super heavy-tailed noise in bandit learning problems: We propose a novel robust statistical estimator, mean of medians, which estimates a random variable by computing the empirical mean of a sequence of empirical medians. We then present a generic reductionist algorithmic framework for solving bandit learning problems (including multi-armed and linear bandit problem): the mean of medians estimator can be applied to nearly any bandit learning algorithm as a black-box filtering for its reward signals and obtain similar regret bound as if the reward is sub-Gaussian. We show that the regret bound is near-optimal even with very heavy-tailed noise. We also empirically demonstrate the effectiveness of the proposed algorithm, which further corroborates our theoretical results.

Pritam Majumder, Jiayi Huang, Sungkeun Kim et al.

The resurgence of near-memory processing (NMP) with the advent of big data has shifted the computation paradigm from processor-centric to memory-centric computing. To meet the bandwidth and capacity demands of memory-centric computing, 3D memory has been adopted to form a scalable memory-cube network. Along with NMP and memory system development, the mapping for placing data and guiding computation in the memory-cube network has become crucial in driving the performance improvement in NMP. However, it is very challenging to design a universal optimal mapping for all applications due to unique application behavior and intractable decision space. In this paper, we propose an artificially intelligent memory mapping scheme, AIMM, that optimizes data placement and resource utilization through page and computation remapping. Our proposed technique involves continuously evaluating and learning the impact of mapping decisions on system performance for any application. AIMM uses a neural network to achieve a near-optimal mapping during execution, trained using a reinforcement learning algorithm that is known to be effective for exploring a vast design space. We also provide a detailed AIMM hardware design that can be adopted as a plugin module for various NMP systems. Our experimental evaluation shows that AIMM improves the baseline NMP performance in single and multiple program scenario by up to 70% and 50%, respectively.

Jiayi Huang, Mostofa Patwary, Gregory Diamos

We show that recent innovations in deep reinforcement learning can effectively color very large graphs -- a well-known NP-hard problem with clear commercial applications. Because the Monte Carlo Tree Search with Upper Confidence Bound algorithm used in AlphaGoZero can improve the performance of a given heuristic, our approach allows deep neural networks trained using high performance computing (HPC) technologies to transform computation into improved heuristics with zero prior knowledge. Key to our approach is the introduction of a novel deep neural network architecture (FastColorNet) that has access to the full graph context and requires $O(V)$ time and space to color a graph with $V$ vertices, which enables scaling to very large graphs that arise in real applications like parallel computing, compilers, numerical solvers, and design automation, among others. As a result, we are able to learn new state of the art heuristics for graph coloring.