Yicheng Zhang

Ruikang Luo, Yaofeng Song, Han Zhao et al.

Accurate vehicle type classification serves a significant role in the intelligent transportation system. It is critical for ruler to understand the road conditions and usually contributive for the traffic light control system to response correspondingly to alleviate traffic congestion. New technologies and comprehensive data sources, such as aerial photos and remote sensing data, provide richer and high-dimensional information. Also, due to the rapid development of deep neural network technology, image based vehicle classification methods can better extract underlying objective features when processing data. Recently, several deep learning models have been proposed to solve the problem. However, traditional pure convolutional based approaches have constraints on global information extraction, and the complex environment, such as bad weather, seriously limits the recognition capability. To improve the vehicle type classification capability under complex environment, this study proposes a novel Densely Connected Convolutional Transformer in Transformer Neural Network (Dense-TNT) framework for the vehicle type classification by stacking Densely Connected Convolutional Network (DenseNet) and Transformer in Transformer (TNT) layers. Three-region vehicle data and four different weather conditions are deployed for recognition capability evaluation. Experimental findings validate the recognition ability of our proposed vehicle classification model with little decay, even under the heavy foggy weather condition.

Ruikang Luo, Yaofeng Song, Liping Huang et al.

Accurate long series forecasting of traffic information is critical for the development of intelligent traffic systems. We may benefit from the rapid growth of neural network analysis technology to better understand the underlying functioning patterns of traffic networks as a result of this progress. Due to the fact that traffic data and facility utilization circumstances are sequentially dependent on past and present situations, several related neural network techniques based on temporal dependency extraction models have been developed to solve the problem. The complicated topological road structure, on the other hand, amplifies the effect of spatial interdependence, which cannot be captured by pure temporal extraction approaches. Additionally, the typical Deep Recurrent Neural Network (RNN) topology has a constraint on global information extraction, which is required for comprehensive long-term prediction. This study proposes a new spatial-temporal neural network architecture, called Spatial-Temporal Graph-Informer (STGIN), to handle the long-term traffic parameters forecasting issue by merging the Informer and Graph Attention Network (GAT) layers for spatial and temporal relationships extraction. The attention mechanism potentially guarantees long-term prediction performance without significant information loss from distant inputs. On two real-world traffic datasets with varying horizons, experimental findings validate the long sequence prediction abilities, and further interpretation is provided.

Ruikang Luo, Yaofeng Song, Liping Huang et al.

Electric Vehicle (EV) charging demand and charging station availability forecasting is one of the challenges in the intelligent transportation system. With the accurate EV station situation prediction, suitable charging behaviors could be scheduled in advance to relieve range anxiety. Many existing deep learning methods are proposed to address this issue, however, due to the complex road network structure and comprehensive external factors, such as point of interests (POIs) and weather effects, many commonly used algorithms could just extract the historical usage information without considering comprehensive influence of external factors. To enhance the prediction accuracy and interpretability, the Attribute-Augmented Spatial-Temporal Graph Informer (AST-GIN) structure is proposed in this study by combining the Graph Convolutional Network (GCN) layer and the Informer layer to extract both external and internal spatial-temporal dependence of relevant transportation data. And the external factors are modeled as dynamic attributes by the attribute-augmented encoder for training. AST-GIN model is tested on the data collected in Dundee City and experimental results show the effectiveness of our model considering external factors influence over various horizon settings compared with other baselines.

Yuxuan Li, Xiang Li, Yunheng Li et al.

With the rapid advancement of remote sensing technology, high-resolution multi-modal imagery is now more widely accessible. Conventional Object detection models are trained on a single dataset, often restricted to a specific imaging modality and annotation format. However, such an approach overlooks the valuable shared knowledge across multi-modalities and limits the model's applicability in more versatile scenarios. This paper introduces a new task called Multi-Modal Datasets and Multi-Task Object Detection (M2Det) for remote sensing, designed to accurately detect horizontal or oriented objects from any sensor modality. This task poses challenges due to 1) the trade-offs involved in managing multi-modal modelling and 2) the complexities of multi-task optimization. To address these, we establish a benchmark dataset and propose a unified model, SM3Det (Single Model for Multi-Modal datasets and Multi-Task object Detection). SM3Det leverages a grid-level sparse MoE backbone to enable joint knowledge learning while preserving distinct feature representations for different modalities. Furthermore, it integrates a consistency and synchronization optimization strategy using dynamic learning rate adjustment, allowing it to effectively handle varying levels of learning difficulty across modalities and tasks. Extensive experiments demonstrate SM3Det's effectiveness and generalizability, consistently outperforming specialized models on individual datasets. The code is available at https://github.com/zcablii/SM3Det.

Wenti Yin, Huaxin Zhang, Xiang Wang et al.

Recent advancements in weakly-supervised video anomaly detection have achieved remarkable performance by applying the multiple instance learning paradigm based on multimodal foundation models such as CLIP to highlight anomalous instances and classify categories. However, their objectives may tend to detect the most salient response segments, while neglecting to mine diverse normal patterns separated from anomalies, and are prone to category confusion due to similar appearance, leading to unsatisfactory fine-grained classification results. Therefore, we propose a novel Disentangled Semantic Alignment Network (DSANet) to explicitly separate abnormal and normal features from coarse-grained and fine-grained aspects, enhancing the distinguishability. Specifically, at the coarse-grained level, we introduce a self-guided normality modeling branch that reconstructs input video features under the guidance of learned normal prototypes, encouraging the model to exploit normality cues inherent in the video, thereby improving the temporal separation of normal patterns and anomalous events. At the fine-grained level, we present a decoupled contrastive semantic alignment mechanism, which first temporally decomposes each video into event-centric and background-centric components using frame-level anomaly scores and then applies visual-language contrastive learning to enhance class-discriminative representations. Comprehensive experiments on two standard benchmarks, namely XD-Violence and UCF-Crime, demonstrate that DSANet outperforms existing state-of-the-art methods.

Yuxuan Li, Yicheng Zhang, Wenhao Tang et al.

Modern computer vision is converging on a closed loop in which perception, reasoning and generation mutually reinforce each other. However, this loop remains incomplete: the top-down influence of high-level reasoning on the foundational learning of low-level perceptual features is not yet underexplored. This paper addresses this gap by proposing a new paradigm for pretraining foundation models in downstream domains. We introduce Visual insTruction Pretraining (ViTP), a novel approach that directly leverages reasoning to enhance perception. ViTP embeds a Vision Transformer (ViT) backbone within a Vision-Language Model and pretrains it end-to-end using a rich corpus of visual instruction data curated from target downstream domains. ViTP is powered by our proposed Visual Robustness Learning (VRL), which compels the ViT to learn robust and domain-relevant features from a sparse set of visual tokens. Extensive experiments on 16 challenging remote sensing and medical imaging benchmarks demonstrate that ViTP establishes new state-of-the-art performance across a diverse range of downstream tasks. The code is available at https://github.com/zcablii/ViTP.

Tianyu Yu, Kechen Fang, Zihao Wan et al.

Most Vision Language Models (VLMs) directly map outputs from ViT encoders to the LLM via a lightweight projector. While effective, recent analysis suggests this architecture suffers from an alignment challenge: visual features remain distant from the text space in the initial layers of the LLM, forcing the model to waste critical depth~\cite{zhang-etal-2024-investigating,artzy-schwartz-2024-attend} on superficial modality alignment rather than deep understanding and complex reasoning. In this work, we propose Deep Pre-Alignment (DPA), a novel architecture that replaces the standard ViT encoder with a small VLM as perceiver, ensuring visual features are deeply aligned with the text space of the target large language model. Comprehensive experiments demonstrate the effectiveness of DPA. On the 4B parameter scale, DPA outperforms baselines by 1.9 points across 8 multimodal benchmarks, with gains widening to 3.0 points at the 32B scale. Moreover, by offloading alignment to the perceiver, DPA achieves a 32.9\% reduction in language capability forgetting over 3 text benchmarks. We further demonstrate that these gains are consistent across different LLM families including Qwen3 and LLaMA 3.2, highlighting the generality of our approach. Beyond performance, DPA also offers a seamless upgrade path for current VLM development, requiring only a modular replacement for the visual encoder with marginal computation overhead.

Yicheng Zhang, Zhen Qin, Zhaomin Wu et al.

Retrieval-augmented generation (RAG) enables large language models (LLMs) to produce evidence-based responses, and its performance hinges on the matching between the retriever and LLMs. Retriever optimization has emerged as an efficient alternative to fine-tuning LLMs. However, existing solutions suffer from objective mismatch between retriever optimization and the goal of RAG pipeline. Reinforcement learning (RL) provides a promising solution to address this limitation, yet applying RL to retriever optimization introduces two fundamental challenges: 1) the deterministic retrieval is incompatible with RL formulations, and 2) state aliasing arises from query-only retrieval in multi-hop reasoning. To address these challenges, we replace deterministic retrieval with stochastic sampling and formulate RAG as a Markov decision process, making retriever optimizable by RL. Further, we incorporate retrieval history into the state at each retrieval step to mitigate state aliasing. Extensive experiments across diverse RAG pipelines, datasets, and retriever scales demonstrate consistent improvements of our approach in RAG performance.

Bohan Lyu, Yucheng Yang, Siqiao Huang et al.

Modern AI progress has been driven by ML methods that are generalizable across settings and scalable to larger regimes. As large language models demonstrate advanced capabilities in reasoning, coding, and engineering tasks, it is increasingly important to understand whether they can discover such methods rather than only apply existing ones. We introduce MLS-Bench, a benchmark for evaluating whether AI systems can invent generalizable and scalable ML methods. MLS-Bench contains 140 tasks across 12 domains, each requiring an agent to improve one targeted component of an ML system or algorithm and demonstrate that the improvement generalizes across controlled settings and scales. We find that current agents remain far from reliably surpassing human-designed methods, and that engineering-style tuning is easier for them than genuine method invention. We further study the effects of test-time scaling, adaptive compute allocation, and context provision on agents' discovery performance, together with case studies of their behavior. Our analyses suggest that the bottleneck is not only in proposing new methods, but also in the scientific insight needed to plan, validate, and scale claims about them. More search, compute, or context alone does not remove this bottleneck. We build and maintain a community platform for cumulative and comparable iteration, and release the data and code at https://mls-bench.com.

Yicheng Zhang, Zhen Qin, Zhaomin Wu et al.

Large language models (LLMs) are increasingly powering web-based applications, whose effectiveness relies on fine-tuning with large-scale instruction data. However, such data often contains valuable or sensitive information that limits its public sharing among business organizations. Federated learning (FL) enables collaborative fine-tuning of LLMs without accessing raw data. Existing approaches to federated LLM fine-tuning usually adopt a uniform model architecture, making it challenging to fit highly heterogeneous client-side data in varying domains and tasks, e.g., hospitals and financial institutions conducting federated fine-tuning may require different LLM architectures due to the distinct nature of their domains and tasks. To address this, we propose FedAMoLE, a lightweight personalized FL framework that enables data-driven heterogeneous model architectures. It features a heterogeneous mixture of low-rank adaptation (LoRA) experts module to aggregate architecturally heterogeneous models and a reverse selection-based expert assignment strategy to tailor model architectures for each client based on data distributions. Experiments across seven scenarios demonstrate that FedAMoLE improves client-side performance by an average of 5.97% over existing approaches while maintaining practical memory, communication, and computation overhead.

Liping Huang, Sheng Zhang, Yicheng Zhang et al.

Aircraft landing time (ALT) prediction is crucial for air traffic management, especially for arrival aircraft sequencing on the runway. In this study, a trajectory image-based deep learning method is proposed to predict ALTs for the aircraft entering the research airspace that covers the Terminal Maneuvering Area (TMA). Specifically, the trajectories of all airborne arrival aircraft within the temporal capture window are used to generate an image with the target aircraft trajectory labeled as red and all background aircraft trajectory labeled as blue. The trajectory images contain various information, including the aircraft position, speed, heading, relative distances, and arrival traffic flows. It enables us to use state-of-the-art deep convolution neural networks for ALT modeling. We also use real-time runway usage obtained from the trajectory data and the external information such as aircraft types and weather conditions as additional inputs. Moreover, a convolution neural network (CNN) based module is designed for automatic holding-related featurizing, which takes the trajectory images, the leading aircraft holding status, and their time and speed gap at the research airspace boundary as its inputs. Its output is further fed into the final end-to-end ALT prediction. The proposed ALT prediction approach is applied to Singapore Changi Airport (ICAO Code: WSSS) using one-month Automatic Dependent Surveillance-Broadcast (ADS-B) data from November 1 to November 30, 2022. Experimental results show that by integrating the holding featurization, we can reduce the mean absolute error (MAE) from 82.23 seconds to 43.96 seconds, and achieve an average accuracy of 96.1\%, with 79.4\% of the predictions errors being less than 60 seconds.

Jialong Zuo, Haoyou Deng, Hanyu Zhou et al.

The rapid evolution of text-to-image generation models has revolutionized visual content creation. While commercial products like Nano Banana Pro have garnered significant attention, their potential as generalist solvers for traditional low-level vision challenges remains largely underexplored. In this study, we investigate the critical question: Is Nano Banana Pro a Low-Level Vision All-Rounder? We conducted a comprehensive zero-shot evaluation across 14 distinct low-level tasks spanning 40 diverse datasets. By utilizing simple textual prompts without fine-tuning, we benchmarked Nano Banana Pro against state-of-the-art specialist models. Our extensive analysis reveals a distinct performance dichotomy: while \textbf{Nano Banana Pro demonstrates superior subjective visual quality}, often hallucinating plausible high-frequency details that surpass specialist models, it lags behind in traditional reference-based quantitative metrics. We attribute this discrepancy to the inherent stochasticity of generative models, which struggle to maintain the strict pixel-level consistency required by conventional metrics. This report identifies Nano Banana Pro as a capable zero-shot contender for low-level vision tasks, while highlighting that achieving the high fidelity of domain specialists remains a significant hurdle.

Liping Huang, Yicheng Zhang, Yifang Yin et al.

Estimated time of arrival (ETA) for airborne aircraft in real-time is crucial for arrival management in aviation, particularly for runway sequencing. Given the rapidly changing airspace context, the ETA prediction efficiency is as important as its accuracy in a real-time arrival aircraft management system. In this study, we utilize a feature tokenization-based Transformer model to efficiently predict aircraft ETA. Feature tokenization projects raw inputs to latent spaces, while the multi-head self-attention mechanism in the Transformer captures important aspects of the projections, alleviating the need for complex feature engineering. Moreover, the Transformer's parallel computation capability allows it to handle ETA requests at a high frequency, i.e., 1HZ, which is essential for a real-time arrival management system. The model inputs include raw data, such as aircraft latitude, longitude, ground speed, theta degree for the airport, day and hour from track data, the weather context, and aircraft wake turbulence category. With a data sampling rate of 1HZ, the ETA prediction is updated every second. We apply the proposed aircraft ETA prediction approach to Singapore Changi Airport (ICAO Code: WSSS) using one-month Automatic Dependent Surveillance-Broadcast (ADS-B) data from October 1 to October 31, 2022. In the experimental evaluation, the ETA modeling covers all aircraft within a range of 10NM to 300NM from WSSS. The results show that our proposed method method outperforms the commonly used boosting tree based model, improving accuracy by 7\% compared to XGBoost, while requiring only 39\% of its computing time. Experimental results also indicate that, with 40 aircraft in the airspace at a given timestamp, the ETA inference time is only 51.7 microseconds, making it promising for real-time arrival management systems.

Binjie Guo, Hanyu Zheng, Haohan Jiang et al.

Due to the lack of a method to efficiently represent the multimodal information of a protein, including its structure and sequence information, predicting compound-protein binding affinity (CPA) still suffers from low accuracy when applying machine learning methods. To overcome this limitation, in a novel end-to-end architecture (named FeatNN), we develop a coevolutionary strategy to jointly represent the structure and sequence features of proteins and ultimately optimize the mathematical models for predicting CPA. Furthermore, from the perspective of data-driven approach, we proposed a rational method that can utilize both high- and low-quality databases to optimize the accuracy and generalization ability of FeatNN in CPA prediction tasks. Notably, we visually interpret the feature interaction process between sequence and structure in the rationally designed architecture. As a result, FeatNN considerably outperforms the state-of-the-art (SOTA) baseline in virtual drug screening tasks, indicating the feasibility of this approach for practical use. FeatNN provides an outstanding method for higher CPA prediction accuracy and better generalization ability by efficiently representing multimodal information of proteins via a coevolutionary strategy.

Yicheng Zhang, Qiang Ling

As a major type of transportation equipments, bicycles, including electrical bicycles, are distributed almost everywhere in China. The accidents caused by bicycles have become a serious threat to the public safety. So bicycle detection is one major task of traffic video surveillance systems in China. In this paper, a method based on multi-feature and multi-frame fusion is presented for bicycle detection in low-resolution traffic videos. It first extracts some geometric features of objects from each frame image, then concatenate multiple features into a feature vector and use linear support vector machine (SVM) to learn a classifier, or put these features into a cascade classifier, to yield a preliminary detection result regarding whether an object is a bicycle. It further fuses these preliminary detection results from multiple frames to provide a more reliable detection decision, together with a confidence level of that decision. Experimental results show that this method based on multi-feature and multi-frame fusion can identify bicycles with high accuracy and low computational complexity. It is, therefore, applicable for real-time traffic video surveillance systems.