Yukun Yang

Jiarun Liu, Daguang Jiang, Yukun Yang et al.

Deep learning has made many remarkable achievements in many fields but suffers from noisy labels in datasets. The state-of-the-art learning with noisy label method Co-teaching and Co-teaching+ confronts the noisy label by mutual-information between dual-network. However, the dual network always tends to convergent which would weaken the dual-network mechanism to resist the noisy labels. In this paper, we proposed a noise-tolerant framework named MLC in an end-to-end manner. It adjusts the dual-network with divergent regularization to ensure the effectiveness of the mechanism. In addition, we correct the label distribution according to the agreement between dual-networks. The proposed method can utilize the noisy data to improve the accuracy, generalization, and robustness of the network. We test the proposed method on the simulate noisy dataset MNIST, CIFAR-10, and the real-world noisy dataset Clothing1M. The experimental result shows that our method outperforms the previous state-of-the-art method. Besides, our method is network-free thus it is applicable to many tasks. Our code can be found at https://github.com/JiarunLiu/MLC.

Peng Kuai, Yukun Yang, Shaolun Ruan et al.

Rare disease diagnosis is inherently challenging due to heterogeneous symptoms, limited clinical familiarity, and fragmented evidence across specialties. Recent large language model (LLM)-based agentic systems have shown promise by simulating multidisciplinary team discussions to generate and evaluate diagnostic hypotheses. However, fully automated diagnosis remains unrealistic, and existing human-in-the-loop approaches provide limited support for effective clinician-agent collaboration. In practice, clinicians are often presented with final diagnostic outputs and lengthy, unstructured agent discussion logs, making it difficult to inspect reasoning, intervene in a timely manner, or guide agent deliberation effectively. To address these challenges, we developed MDTRoom, an interactive system that transforms multi-agent discussions from linear transcripts into a structured, inspectable workspace. The system externalizes patient data, evidence provenance, hypothesis evolution, and inter-agent conflicts as interconnected visual objects, enabling clinicians to efficiently examine, intervene in, and guide agent reasoning. Our evaluation demonstrates the effectiveness of MDTRoom in supporting clinician-agent collaboration.

Yukun Yang, Peng Li

Gradient-based first-order adaptive optimization methods such as the Adam optimizer are prevalent in training artificial networks, achieving the state-of-the-art results. This work attempts to answer the question whether it is viable for biological neural systems to adopt such optimization methods. To this end, we demonstrate a realization of the Adam optimizer using biologically-plausible mechanisms in synapses. The proposed learning rule has clear biological correspondence, runs continuously in time, and achieves performance to comparable Adam's. In addition, we present a new approach, inspired by the predisposition property of synapses observed in neuroscience, to circumvent the biological implausibility of the weight transport problem in backpropagation (BP). With only local information and no separate training phases, this method establishes and maintains weight symmetry in the forward and backward signaling paths, and is applicable to the proposed biologically plausible Adam learning rule. These mechanisms may shed light on the way in which biological synaptic dynamics facilitate learning.

Yukun Yang

This report explores the theory that explains the high sparsity phenomenon \citep{tosato2023emergent} observed in the forward-forward algorithm \citep{hinton2022forward}. The two theorems proposed predict the sparsity changes of a single data point's activation in two cases: Theorem \ref{theorem:1}: Decrease the goodness of the whole batch. Theorem \ref{theorem:2}: Apply the complete forward forward algorithm to decrease the goodness for negative data and increase the goodness for positive data. The theory aligns well with the experiments tested on the MNIST dataset.

Junjie Zhang, Jingyi Xi, Zhuoyang Song et al.

Training large language models (LLMs) to act as autonomous agents for multi-turn, long-horizon tasks remains significant challenges in scalability and training efficiency. To address this, we introduce L-Zero (L0), a scalable, end-to-end training pipeline for general-purpose agents. Featuring a low-cost, extensible, and sandboxed concurrent agent worker pool, L0 lowers the barrier for applying reinforcement learning in complex environments. We also introduce NB-Agent, the agent scaffold within L0, which operates in a "code-as-action" fashion via a Read-Eval-Print-Loop (REPL). We evaluate L0 on factuality question-answering benchmarks. Our experiments demonstrate that a base model can develop robust problem-solving skills using solely Reinforcement Learning with Verifiable Rewards (RLVR). On the Qwen2.5-7B-Instruct model, our method boosts accuracy on SimpleQA from 30 % to 80 % and on HotpotQA from 22 % to 41 %. We have open-sourced the entire L0 system, including our L0 series models, the NB-Agent, a complete training pipeline, and the corresponding training recipes on (https://github.com/cmriat/l0).

Yukun Yang, Peng Li

Several recent studies attempt to address the biological implausibility of the well-known backpropagation (BP) method. While promising methods such as feedback alignment, direct feedback alignment, and their variants like sign-concordant feedback alignment tackle BP's weight transport problem, their validity remains controversial owing to a set of other unsolved issues. In this work, we answer the question of whether it is possible to realize random backpropagation solely based on mechanisms observed in neuroscience. We propose a hypothetical framework consisting of a new microcircuit architecture and its supporting Hebbian learning rules. Comprising three types of cells and two types of synaptic connectivity, the proposed microcircuit architecture computes and propagates error signals through local feedback connections and supports the training of multi-layered spiking neural networks with a globally defined spiking error function. We employ the Hebbian rule operating in local compartments to update synaptic weights and achieve supervised learning in a biologically plausible manner. Finally, we interpret the proposed framework from an optimization point of view and show its equivalence to sign-concordant feedback alignment. The proposed framework is benchmarked on several datasets including MNIST and CIFAR10, demonstrating promising BP-comparable accuracy.

Bin Ren, Hang Guo, Lei Sun et al.

This paper presents a comprehensive review of the NTIRE 2025 Challenge on Single-Image Efficient Super-Resolution (ESR). The challenge aimed to advance the development of deep models that optimize key computational metrics, i.e., runtime, parameters, and FLOPs, while achieving a PSNR of at least 26.90 dB on the $\operatorname{DIV2K\_LSDIR\_valid}$ dataset and 26.99 dB on the $\operatorname{DIV2K\_LSDIR\_test}$ dataset. A robust participation saw \textbf{244} registered entrants, with \textbf{43} teams submitting valid entries. This report meticulously analyzes these methods and results, emphasizing groundbreaking advancements in state-of-the-art single-image ESR techniques. The analysis highlights innovative approaches and establishes benchmarks for future research in the field.

Charles Chiang, Simret Gebreegziabher, Annalisa Szymanski et al.

LLM-as-a-judge approaches have emerged as a scalable solution for evaluating model behaviors, yet they rely on evaluation criteria often created by a single individual, embedding that person's assumptions, priorities, and interpretive lens. In practice, defining such criteria is a collaborative and contested process involving multiple stakeholders with different values, interpretations, and priorities; an aspect largely unsupported by existing tools. To examine this problem in depth, we present a formative study examining how stakeholders collaboratively create, negotiate, and refine evaluation criteria for LLM-as-a-judge systems. Our findings reveal challenges in human oversight, including difficulties in establishing shared understanding, aligning values across stakeholders with different expertise and priorities, and translating nuanced human judgments into criteria that are interpretable and actionable for LLM judges. Based on these insights, we developed MultEval, a system that supports collaborative criteria by enabling multiple evaluators to surface and diagnose disagreements using consensus-building theory, iteratively revise criteria with attached examples and proposal history, and maintain transparency over how judgments are encoded into an automated evaluator. We further report a case study in which a team of domain experts used MultEval to collaboratively author criteria, illustrating how coordination and collaborative consensus-making shape criteria evolution.

Yukun Yang

Depression is a major global public health challenge and its early identification is crucial. Social media data provides a new perspective for depression detection, but existing methods face limitations such as insufficient accuracy, insufficient utilization of time series features, and high annotation costs. To this end, this study proposes the LMILAtt model, which innovatively integrates Long Short-Term Memory autoencoders and attention mechanisms: firstly, the temporal dynamic features of user tweets (such as depressive tendency evolution patterns) are extracted through unsupervised LSTM autoencoders. Secondly, the attention mechanism is used to dynamically weight key texts (such as early depression signals) and construct a multi-example learning architecture to improve the accuracy of user-level detection. Finally, the performance was verified on the WU3D dataset labeled by professional medicine. Experiments show that the model is significantly better than the baseline model in terms of accuracy, recall and F1 score. In addition, the weakly supervised learning strategy significantly reduces the cost of labeling and provides an efficient solution for large-scale social media depression screening.

Yukun Yang

We introduce an end-to-end pipeline that leverages a fine-tuned YOLO12l-seg model -- trained on over 500 annotated post-blast images -- to deliver real-time instance segmentation (Box mAP@0.5 ~ 0.769, Mask mAP@0.5 ~ 0.800 at ~ 15 FPS). High-fidelity masks are converted into normalized 3D coordinates, from which we extract multi-metric spatial descriptors: principal component directions, kernel density hotspots, size-depth regression, and Delaunay edge statistics. We present four representative examples to illustrate key fragmentation patterns. Experimental results confirm the framework's accuracy, robustness to small-object crowding, and feasibility for rapid, automated blast-effect assessment in field conditions.

Yukun Yang, Naihao Wang, Haixin Yang et al.

Label noise is a common issue in real-world datasets that inevitably impacts the generalization of models. This study focuses on robust classification tasks where the label noise is instance-dependent. Estimating the transition matrix accurately in this task is challenging, and methods based on sample selection often exhibit confirmation bias to varying degrees. Sparse over-parameterized training (SOP) has been theoretically effective in estimating and recovering label noise, offering a novel solution for noise-label learning. However, this study empirically observes and verifies a technical flaw of SOP: the lack of coordination between model predictions and noise recovery leads to increased generalization error. To address this, we propose a method called Coordinated Sparse Recovery (CSR). CSR introduces a collaboration matrix and confidence weights to coordinate model predictions and noise recovery, reducing error leakage. Based on CSR, this study designs a joint sample selection strategy and constructs a comprehensive and powerful learning framework called CSR+. CSR+ significantly reduces confirmation bias, especially for datasets with more classes and a high proportion of instance-specific noise. Experimental results on simulated and real-world noisy datasets demonstrate that both CSR and CSR+ achieve outstanding performance compared to methods at the same level.

Yukun Yang, Peng Li

Our brain consists of biological neurons encoding information through accurate spike timing, yet both the architecture and learning rules of our brain remain largely unknown. Comparing to the recent development of backpropagation-based (BP-based) methods that are able to train spiking neural networks (SNNs) with high accuracy, biologically plausible methods are still in their infancy. In this work, we wish to answer the question of whether it is possible to attain comparable accuracy of SNNs trained by BP-based rules with bio-plausible mechanisms. We propose a new bio-plausible learning framework, consisting of two components: a new architecture, and its supporting learning rules. With two types of cells and four types of synaptic connections, the proposed local microcircuit architecture can compute and propagate error signals through local feedback connections and support training of multi-layers SNNs with a globally defined spiking error function. Under our microcircuit architecture, we employ the Spike-Timing-Dependent-Plasticity (STDP) rule operating in local compartments to update synaptic weights and achieve supervised learning in a biologically plausible manner. Finally, We interpret the proposed framework from an optimization point of view and show the equivalence between it and the BP-based rules under a special circumstance. Our experiments show that the proposed framework demonstrates learning accuracy comparable to BP-based rules and may provide new insights on how learning is orchestrated in biological systems.

Yukun Yang, Wenrui Zhang, Peng Li

While backpropagation (BP) has been applied to spiking neural networks (SNNs) achieving encouraging results, a key challenge involved is to backpropagate a continuous-valued loss over layers of spiking neurons exhibiting discontinuous all-or-none firing activities. Existing methods deal with this difficulty by introducing compromises that come with their own limitations, leading to potential performance degradation. We propose a novel BP-like method, called neighborhood aggregation (NA), which computes accurate error gradients guiding weight updates that may lead to discontinuous modifications of firing activities. NA achieves this goal by aggregating finite differences of the loss over multiple perturbed membrane potential waveforms in the neighborhood of the present membrane potential of each neuron while utilizing a new membrane potential distance function. Our experiments show that the proposed NA algorithm delivers the state-of-the-art performance for SNN training on several datasets.

Yukun Yang

Spiking neural networks (SNN) are usually more energy-efficient as compared to Artificial neural networks (ANN), and the way they work has a great similarity with our brain. Back-propagation (BP) has shown its strong power in training ANN in recent years. However, since spike behavior is non-differentiable, BP cannot be applied to SNN directly. Although prior works demonstrated several ways to approximate the BP-gradient in both spatial and temporal directions either through surrogate gradient or randomness, they omitted the temporal dependency introduced by the reset mechanism between each step. In this article, we target on theoretical completion and investigate the effect of the missing term thoroughly. By adding the temporal dependency of the reset mechanism, the new algorithm is more robust to learning-rate adjustments on a toy dataset but does not show much improvement on larger learning tasks like CIFAR-10. Empirically speaking, the benefits of the missing term are not worth the additional computational overhead. In many cases, the missing term can be ignored.

Ximing Qiao, Yukun Yang, Hai Li

Neural backdoor attack is emerging as a severe security threat to deep learning, while the capability of existing defense methods is limited, especially for complex backdoor triggers. In the work, we explore the space formed by the pixel values of all possible backdoor triggers. An original trigger used by an attacker to build the backdoored model represents only a point in the space. It then will be generalized into a distribution of valid triggers, all of which can influence the backdoored model. Thus, previous methods that model only one point of the trigger distribution is not sufficient. Getting the entire trigger distribution, e.g., via generative modeling, is a key to effective defense. However, existing generative modeling techniques for image generation are not applicable to the backdoor scenario as the trigger distribution is completely unknown. In this work, we propose max-entropy staircase approximator (MESA), an algorithm for high-dimensional sampling-free generative modeling and use it to recover the trigger distribution. We also develop a defense technique to remove the triggers from the backdoored model. Our experiments on Cifar10/100 dataset demonstrate the effectiveness of MESA in modeling the trigger distribution and the robustness of the proposed defense method.

Hsin-Pai Cheng, Tunhou Zhang, Yukun Yang et al.

Designing neural architectures for edge devices is subject to constraints of accuracy, inference latency, and computational cost. Traditionally, researchers manually craft deep neural networks to meet the needs of mobile devices. Neural Architecture Search (NAS) was proposed to automate the neural architecture design without requiring extensive domain expertise and significant manual efforts. Recent works utilized NAS to design mobile models by taking into account hardware constraints and achieved state-of-the-art accuracy with fewer parameters and less computational cost measured in Multiply-accumulates (MACs). To find highly compact neural architectures, existing works relies on predefined cells and directly applying width multiplier, which may potentially limit the model flexibility, reduce the useful feature map information, and cause accuracy drop. To conquer this issue, we propose GRAM(GRAph propagation as Meta-knowledge) that adopts fine-grained (node-wise) search method and accumulates the knowledge learned in updates into a meta-graph. As a result, GRAM can enable more flexible search space and achieve higher search efficiency. Without the constraints of predefined cell or blocks, we propose a new structure-level pruning method to remove redundant operations in neural architectures. SwiftNet, which is a set of models discovered by GRAM, outperforms MobileNet-V2 by 2.15x higher accuracy density and 2.42x faster with similar accuracy. Compared with FBNet, SwiftNet reduces the search cost by 26x and achieves 2.35x higher accuracy density and 1.47x speedup while preserving similar accuracy. SwiftNetcan obtain 63.28% top-1 accuracy on ImageNet-1K with only 53M MACs and 2.07M parameters. The corresponding inference latency is only 19.09 ms on Google Pixel 1.