Yijing Yang

Yijing Yang, Hongyu Fu, C. -C. Jay Kuo

The design of robust learning systems that offer stable performance under a wide range of supervision degrees is investigated in this work. We choose the image classification problem as an illustrative example and focus on the design of modularized systems that consist of three learning modules: representation learning, feature learning and decision learning. We discuss ways to adjust each module so that the design is robust with respect to different training sample numbers. Based on these ideas, we propose two families of learning systems. One adopts the classical histogram of oriented gradients (HOG) features while the other uses successive-subspace-learning (SSL) features. We test their performance against LeNet-5, which is an end-to-end optimized neural network, for MNIST and Fashion-MNIST datasets. The number of training samples per image class goes from the extremely weak supervision condition (i.e., 1 labeled sample per class) to the strong supervision condition (i.e., 4096 labeled sample per class) with gradual transition in between (i.e., $2^n$, $n=0, 1, \cdots, 12$). Experimental results show that the two families of modularized learning systems have more robust performance than LeNet-5. They both outperform LeNet-5 by a large margin for small $n$ and have performance comparable with that of LeNet-5 for large $n$.

Hongyu Fu, Yijing Yang, Yuhuai Liu et al.

Built upon the decision tree (DT) classification and regression idea, the subspace learning machine (SLM) has been recently proposed to offer higher performance in general classification and regression tasks. Its performance improvement is reached at the expense of higher computational complexity. In this work, we investigate two ways to accelerate SLM. First, we adopt the particle swarm optimization (PSO) algorithm to speed up the search of a discriminant dimension that is expressed as a linear combination of current dimensions. The search of optimal weights in the linear combination is computationally heavy. It is accomplished by probabilistic search in original SLM. The acceleration of SLM by PSO requires 10-20 times fewer iterations. Second, we leverage parallel processing in the SLM implementation. Experimental results show that the accelerated SLM method achieves a speed up factor of 577 in training time while maintaining comparable classification/regression performance of original SLM.

Vasileios Magoulianitis, Yijing Yang, C. -C. Jay Kuo

A high-performance unsupervised nuclei instance segmentation (HUNIS) method is proposed in this work. HUNIS consists of two-stage block-wise operations. The first stage includes: 1) adaptive thresholding of pixel intensities, 2) incorporation of nuclei size/shape priors and 3) removal of false positive nuclei instances. Then, HUNIS conducts the second stage segmentation by receiving guidance from the first one. The second stage exploits the segmentation masks obtained in the first stage and leverages color and shape distributions for a more accurate segmentation. The main purpose of the two-stage design is to provide pixel-wise pseudo-labels from the first to the second stage. This self-supervision mechanism is novel and effective. Experimental results on the MoNuSeg dataset show that HUNIS outperforms all other unsupervised methods by a substantial margin. It also has a competitive standing among state-of-the-art supervised methods.

Yijing Yang, Wei Wang, Hongyu Fu et al.

The application of machine learning to image and video data often yields a high dimensional feature space. Effective feature selection techniques identify a discriminant feature subspace that lowers computational and modeling costs with little performance degradation. A novel supervised feature selection methodology is proposed for machine learning decisions in this work. The resulting tests are called the discriminant feature test (DFT) and the relevant feature test (RFT) for the classification and regression problems, respectively. The DFT and RFT procedures are described in detail. Furthermore, we compare the effectiveness of DFT and RFT with several classic feature selection methods. To this end, we use deep features obtained by LeNet-5 for MNIST and Fashion-MNIST datasets as illustrative examples. Other datasets with handcrafted and gene expressions features are also included for performance evaluation. It is shown by experimental results that DFT and RFT can select a lower dimensional feature subspace distinctly and robustly while maintaining high decision performance.

Ailin Huang, Ang Li, Aobo Kong et al.

We introduce Step 3.5 Flash, a sparse Mixture-of-Experts (MoE) model that bridges frontier-level agentic intelligence and computational efficiency. We focus on what matters most when building agents: sharp reasoning and fast, reliable execution. Step 3.5 Flash pairs a 196B-parameter foundation with 11B active parameters for efficient inference. It is optimized with interleaved 3:1 sliding-window/full attention and Multi-Token Prediction (MTP-3) to reduce the latency and cost of multi-round agentic interactions. To reach frontier-level intelligence, we design a scalable reinforcement learning framework that combines verifiable signals with preference feedback, while remaining stable under large-scale off-policy training, enabling consistent self-improvement across mathematics, code, and tool use. Step 3.5 Flash demonstrates strong performance across agent, coding, and math tasks, achieving 85.4% on IMO-AnswerBench, 86.4% on LiveCodeBench-v6 (2024.08-2025.05), 88.2% on tau2-Bench, 69.0% on BrowseComp (with context management), and 51.0% on Terminal-Bench 2.0, comparable to frontier models such as GPT-5.2 xHigh and Gemini 3.0 Pro. By redefining the efficiency frontier, Step 3.5 Flash provides a high-density foundation for deploying sophisticated agents in real-world industrial environments.

Hongyu Fu, Yijing Yang, Vinod K. Mishra et al.

Inspired by the feedforward multilayer perceptron (FF-MLP), decision tree (DT) and extreme learning machine (ELM), a new classification model, called the subspace learning machine (SLM), is proposed in this work. SLM first identifies a discriminant subspace, $S^0$, by examining the discriminant power of each input feature. Then, it uses probabilistic projections of features in $S^0$ to yield 1D subspaces and finds the optimal partition for each of them. This is equivalent to partitioning $S^0$ with hyperplanes. A criterion is developed to choose the best $q$ partitions that yield $2q$ partitioned subspaces among them. We assign $S^0$ to the root node of a decision tree and the intersections of $2q$ subspaces to its child nodes of depth one. The partitioning process is recursively applied at each child node to build an SLM tree. When the samples at a child node are sufficiently pure, the partitioning process stops and each leaf node makes a prediction. The idea can be generalized to regression, leading to the subspace learning regressor (SLR). Furthermore, ensembles of SLM/SLR trees can yield a stronger predictor. Extensive experiments are conducted for performance benchmarking among SLM/SLR trees, ensembles and classical classifiers/regressors.

Yijing Yang, Vasileios Magoulianitis, Xinyu Wang et al.

A statistical attention localization (SAL) method is proposed to facilitate the object classification task in this work. SAL consists of three steps: 1) preliminary attention window selection via decision statistics, 2) attention map refinement, and 3) rectangular attention region finalization. SAL computes soft-decision scores of local squared windows and uses them to identify salient regions in Step 1. To accommodate object of various sizes and shapes, SAL refines the preliminary result and obtain an attention map of more flexible shape in Step 2. Finally, SAL yields a rectangular attention region using the refined attention map and bounding box regularization in Step 3. As an application, we adopt E-PixelHop, which is an object classification solution based on successive subspace learning (SSL), as the baseline. We apply SAL so as to obtain a cropped-out and resized attention region as an alternative input. Classification results of the whole image as well as the attention region are ensembled to achieve the highest classification accuracy. Experiments on the CIFAR-10 dataset are given to demonstrate the advantage of the SAL-assisted object classification method.

Zhaoyuan Ding, Yijing Yang, Han Shu et al.

Segment Anything Model 2 (SAM 2) serves as a core foundation model in the field of video segmentation. Building upon the original SAM model, it introduces a memory bank mechanism and demonstrates outstanding performance in tasks such as semi-supervised video object segmentation and tracking anything. However, the complex computational characteristics of SAM 2's multi-stage image encoder and memory module have raised the barrier to the model's deployment in practical applications. To address this issue, we propose TinySAM 2, a lightweight video segmentation model that balances performance and efficiency. First, a memory quality management mechanism is introduced to select and retain high-informative historical frames as the memory. In addition, a joint-spatial-temporal token compression is proposed that reduces the memory storage and computational cost. Specifically, average pooling is employed to first compress redundancy tokens in the spatial domain. In the temporal domain, informative tokens are selected across frames in the memory bank based on token-level similarity measurement. Besides, we take RepViT as the lightweight image encoder, which further reduces the model parameters. Extensive experiments on challenging datasets such as DAVIS and SA-V demonstrate that TinySAM 2 achieves 90% of the performance of SAM 2.1, with only 7% memory tokens and 3% training data. This study effectively alleviates the bottlenecks in parameter count, computational load, and deployment costs associated with SAM 2, providing a resource-efficient solution for the widespread application of video segmentation models on devices.

Xinghao Chen, Siwei Li, Yijing Yang et al.

Transformer and its variants have shown great potential for various vision tasks in recent years, including image classification, object detection and segmentation. Meanwhile, recent studies also reveal that with proper architecture design, convolutional networks (ConvNets) also achieve competitive performance with transformers. However, no prior methods have explored to utilize pure convolution to build a Transformer-style Decoder module, which is essential for Encoder-Decoder architecture like Detection Transformer (DETR). To this end, in this paper we explore whether we could build query-based detection and segmentation framework with ConvNets instead of sophisticated transformer architecture. We propose a novel mechanism dubbed InterConv to perform interaction between object queries and image features via convolutional layers. Equipped with the proposed InterConv, we build Detection ConvNet (DECO), which is composed of a backbone and convolutional encoder-decoder architecture. We compare the proposed DECO against prior detectors on the challenging COCO benchmark. Despite its simplicity, our DECO achieves competitive performance in terms of detection accuracy and running speed. Specifically, with the ResNet-18 and ResNet-50 backbone, our DECO achieves $40.5\%$ and $47.8\%$ AP with $66$ and $34$ FPS, respectively. The proposed method is also evaluated on the segment anything task, demonstrating similar performance and higher efficiency. We hope the proposed method brings another perspective for designing architectures for vision tasks. Codes are available at https://github.com/xinghaochen/DECO and https://github.com/mindspore-lab/models/tree/master/research/huawei-noah/DECO.

StepFun, Bin Wang, Bojun Wang et al.

Large language models (LLMs) face low hardware efficiency during decoding, especially for long-context reasoning tasks. This paper introduces Step-3, a 321B-parameter VLM with hardware-aware model-system co-design optimized for minimizing decoding costs. Step-3 innovates in two key dimensions: (1) A novel Multi-Matrix Factorization Attention (MFA) mechanism that significantly reduces both KV cache size and computation while maintaining high attention expressiveness, and (2) Attention-FFN Disaggregation (AFD), a distributed inference system that decouples attention and Feed-Forward Network (FFN) layers into specialized subsystems. This co-design achieves unprecedented cost efficiency: Step-3 significantly reduces theoretical decoding costs compared with models like DeepSeek-V3 and Qwen3 MoE 235B, with the gains widening at longer context. Step-3 achieves low cost while activating 38B parameters per token (more than DeepSeek-V3 and Qwen3 MoE 235B), demonstrating that hardware-aligned attention arithmetic intensity, MoE sparsity, and AFD are critical to cost-effectiveness. We perform a head-to-head comparison with DeepSeek-V3 in its favorable scenarios. Our implementation on Hopper GPUs achieves a decoding throughput of up to 4,039 tokens per second per GPU under 50ms TPOT SLA (4K context, FP8, no MTP). It is higher than DeepSeek-V3's 2,324 in the same setup and sets a new Pareto frontier for LLM decoding.

Vasileios Magoulianitis, Peida Han, Yijing Yang et al.

An unsupervised data-driven nuclei segmentation method for histology images, called CBM, is proposed in this work. CBM consists of three modules applied in a block-wise manner: 1) data-driven color transform for energy compaction and dimension reduction, 2) data-driven binarization, and 3) incorporation of geometric priors with morphological processing. CBM comes from the first letter of the three modules - "Color transform", "Binarization" and "Morphological processing". Experiments on the MoNuSeg dataset validate the effectiveness of the proposed CBM method. CBM outperforms all other unsupervised methods and offers a competitive standing among supervised models based on the Aggregated Jaccard Index (AJI) metric.

Yijing Yang, Vasileios Magoulianitis, C. -C. Jay Kuo

Based on PixelHop and PixelHop++, which are recently developed using the successive subspace learning (SSL) framework, we propose an enhanced solution for object classification, called E-PixelHop, in this work. E-PixelHop consists of the following steps. First, to decouple the color channels for a color image, we apply principle component analysis and project RGB three color channels onto two principle subspaces which are processed separately for classification. Second, to address the importance of multi-scale features, we conduct pixel-level classification at each hop with various receptive fields. Third, to further improve pixel-level classification accuracy, we develop a supervised label smoothing (SLS) scheme to ensure prediction consistency. Forth, pixel-level decisions from each hop and from each color subspace are fused together for image-level decision. Fifth, to resolve confusing classes for further performance boosting, we formulate E-PixelHop as a two-stage pipeline. In the first stage, multi-class classification is performed to get a soft decision for each class, where the top 2 classes with the highest probabilities are called confusing classes. Then,we conduct a binary classification in the second stage. The main contributions lie in Steps 1, 3 and 5.We use the classification of the CIFAR-10 dataset as an example to demonstrate the effectiveness of the above-mentioned key components of E-PixelHop.

Yueru Chen, Yijing Yang, Min Zhang et al.

A semi-supervised learning framework using the feedforward-designed convolutional neural networks (FF-CNNs) is proposed for image classification in this work. One unique property of FF-CNNs is that no backpropagation is used in model parameters determination. Since unlabeled data may not always enhance semi-supervised learning, we define an effective quality score and use it to select a subset of unlabeled data in the training process. We conduct experiments on the MNIST, SVHN, and CIFAR-10 datasets, and show that the proposed semi-supervised FF-CNN solution outperforms the CNN trained by backpropagation (BP-CNN) when the amount of labeled data is reduced. Furthermore, we develop an ensemble system that combines the output decision vectors of different semi-supervised FF-CNNs to boost classification accuracy. The ensemble systems can achieve further performance gains on all three benchmarking datasets.

Yueru Chen, Yijing Yang, Wei Wang et al.

An ensemble method that fuses the output decision vectors of multiple feedforward-designed convolutional neural networks (FF-CNNs) to solve the image classification problem is proposed in this work. To enhance the performance of the ensemble system, it is critical to increasing the diversity of FF-CNN models. To achieve this objective, we introduce diversities by adopting three strategies: 1) different parameter settings in convolutional layers, 2) flexible feature subsets fed into the Fully-connected (FC) layers, and 3) multiple image embeddings of the same input source. Furthermore, we partition input samples into easy and hard ones based on their decision confidence scores. As a result, we can develop a new ensemble system tailored to hard samples to further boost classification accuracy. Experiments are conducted on the MNIST and CIFAR-10 datasets to demonstrate the effectiveness of the ensemble method.