Fan Hu

Haifeng Wang, Hua Wu, Tian Wu et al.

In this report, we introduce ERNIE 5.0, a natively autoregressive foundation model desinged for unified multimodal understanding and generation across text, image, video, and audio. All modalities are trained from scratch under a unified next-group-of-tokens prediction objective, based on an ultra-sparse mixture-of-experts (MoE) architecture with modality-agnostic expert routing. To address practical challenges in large-scale deployment under diverse resource constraints, ERNIE 5.0 adopts a novel elastic training paradigm. Within a single pre-training run, the model learns a family of sub-models with varying depths, expert capacities, and routing sparsity, enabling flexible trade-offs among performance, model size, and inference latency in memory- or time-constrained scenarios. Moreover, we systematically address the challenges of scaling reinforcement learning to unified foundation models, thereby guaranteeing efficient and stable post-training under ultra-sparse MoE architectures and diverse multimodal settings. Extensive experiments demonstrate that ERNIE 5.0 achieves strong and balanced performance across multiple modalities. To the best of our knowledge, among publicly disclosed models, ERNIE 5.0 represents the first production-scale realization of a trillion-parameter unified autoregressive model that supports both multimodal understanding and generation. To facilitate further research, we present detailed visualizations of modality-agnostic expert routing in the unified model, alongside comprehensive empirical analysis of elastic training, aiming to offer profound insights to the community.

Ziyue Wang, Aozhu Chen, Fan Hu et al.

Negation is a common linguistic skill that allows human to express what we do NOT want. Naturally, one might expect video retrieval to support natural-language queries with negation, e.g., finding shots of kids sitting on the floor and not playing with a dog. However, the state-of-the-art deep learning based video retrieval models lack such ability, as they are typically trained on video description datasets such as MSR-VTT and VATEX that lack negated descriptions. Their retrieved results basically ignore the negator in the sample query, incorrectly returning videos showing kids playing with dog. This paper presents the first study on learning to understand negation in video retrieval and make contributions as follows. By re-purposing two existing datasets (MSR-VTT and VATEX), we propose a new evaluation protocol for video retrieval with negation. We propose a learning based method for training a negation-aware video retrieval model. The key idea is to first construct a soft negative caption for a specific training video by partially negating its original caption, and then compute a bidirectionally constrained loss on the triplet. This auxiliary loss is weightedly added to a standard retrieval loss. Experiments on the re-purposed benchmarks show that re-training the CLIP (Contrastive Language-Image Pre-Training) model by the proposed method clearly improves its ability to handle queries with negation. In addition, the model performance on the original benchmarks is also improved.

Yeming Gu, Hui Shu, Fei Kang et al.

Binary code similarity detection (BCSD) is widely used in various binary analysis tasks such as vulnerability search, malware detection, clone detection, and patch analysis. Recent studies have shown that the learning-based binary code embedding models perform better than the traditional feature-based approaches. However, previous studies have not delved deeply into the key factors that affect model performance. In this paper, we design extensive ablation studies to explore these influencing factors. The experimental results have provided us with many new insights. We have made innovations in both code representation and model selection: we propose a novel rich-semantic function representation technique to ensure the model captures the intricate nuances of binary code, and we introduce the first UniLM-based binary code embedding model, named UniASM, which includes two newly designed training tasks to learn representations of binary functions. The experimental results show that UniASM outperforms the state-of-the-art (SOTA) approaches on the evaluation datasets. The average scores of Recall@1 on cross-compilers, cross-optimization-levels, and cross-obfuscations have improved by 12.7%, 8.5%, and 22.3%, respectively, compared to the best of the baseline methods. Besides, in the real-world task of known vulnerability search, UniASM outperforms all the current baselines.

Xirong Li, Aozhu Chen, Ziyue Wang et al.

We summarize our TRECVID 2022 Ad-hoc Video Search (AVS) experiments. Our solution is built with two new techniques, namely Lightweight Attentional Feature Fusion (LAFF) for combining diverse visual / textual features and Bidirectional Negation Learning (BNL) for addressing queries that contain negation cues. In particular, LAFF performs feature fusion at both early and late stages and at both text and video ends to exploit diverse (off-the-shelf) features. Compared to multi-head self attention, LAFF is much more compact yet more effective. Its attentional weights can also be used for selecting fewer features, with the retrieval performance mostly preserved. BNL trains a negation-aware video retrieval model by minimizing a bidirectionally constrained loss per triplet, where a triplet consists of a given training video, its original description and a partially negated description. For video feature extraction, we use pre-trained CLIP, BLIP, BEiT, ResNeXt-101 and irCSN. As for text features, we adopt bag-of-words, word2vec, CLIP and BLIP. Our training data consists of MSR-VTT, TGIF and VATEX that were used in our previous participation. In addition, we automatically caption the V3C1 collection for pre-training. The 2022 edition of the TRECVID benchmark has again been a fruitful participation for the RUCMM team. Our best run, with an infAP of 0.262, is ranked at the second place teamwise.

Fan Hu, Dongqi Wang, Huazhen Huang et al.

Drug discovery is vitally important for protecting human against disease. Target-based screening is one of the most popular methods to develop new drugs in the past several decades. This method efficiently screens candidate drugs inhibiting target protein in vitro, but it often fails due to inadequate activity of the selected drugs in vivo. Accurate computational methods are needed to bridge this gap. Here, we propose a novel graph multi task deep learning model to identify compounds carrying both target inhibitory and cell active (MATIC) properties. On a carefully curated SARS-CoV-2 dataset, the proposed MATIC model shows advantages comparing with traditional method in screening effective compounds in vivo. Next, we explored the model interpretability and found that the learned features for target inhibition (in vitro) or cell active (in vivo) tasks are different with molecular property correlations and atom functional attentions. Based on these findings, we utilized a monte carlo based reinforcement learning generative model to generate novel multi-property compounds with both in vitro and in vivo efficacy, thus bridging the gap between target-based and cell-based drug discovery.

Xinyue Wu, Fan Hu, Shaik Jani Babu et al.

Analog circuit design is a time-consuming, experience-driven task in chip development. Despite advances in AI, developing universal, fast, and stable gate sizing methods for analog circuits remains a significant challenge. Recent approaches combine Large Language Models (LLMs) with heuristic search techniques to enhance generalizability, but they often depend on large model sizes and lack portability across different technology nodes. To overcome these limitations, we propose EasySize, the first lightweight gate sizing framework based on a finetuned Qwen3-8B model, designed for universal applicability across process nodes, design specifications, and circuit topologies. EasySize exploits the varying Ease of Attainability (EOA) of performance metrics to dynamically construct task-specific loss functions, enabling efficient heuristic search through global Differential Evolution (DE) and local Particle Swarm Optimization (PSO) within a feedback-enhanced flow. Although finetuned solely on 350nm node data, EasySize achieves strong performance on 5 operational amplifier (Op-Amp) netlists across 180nm, 45nm, and 22nm technology nodes without additional targeted training, and outperforms AutoCkt, a widely-used Reinforcement Learning based sizing framework, on 86.67\% of tasks with more than 96.67\% of simulation resources reduction. We argue that EasySize can significantly reduce the reliance on human expertise and computational resources in gate sizing, thereby accelerating and simplifying the analog circuit design process. EasySize will be open-sourced at a later date.

Ke Liu, Fan Hu, Hui Lin et al.

This paper explores the optimization of Ground Delay Programs (GDP), a prevalent Traffic Management Initiative used in Air Traffic Management (ATM) to reconcile capacity and demand discrepancies at airports. Employing Reinforcement Learning (RL) to manage the inherent uncertainties in the national airspace system-such as weather variability, fluctuating flight demands, and airport arrival rates-we developed two RL models: Behavioral Cloning (BC) and Conservative Q-Learning (CQL). These models are designed to enhance GDP efficiency by utilizing a sophisticated reward function that integrates ground and airborne delays and terminal area congestion. We constructed a simulated single-airport environment, SAGDP_ENV, which incorporates real operational data along with predicted uncertainties to facilitate realistic decision-making scenarios. Utilizing the whole year 2019 data from Newark Liberty International Airport (EWR), our models aimed to preemptively set airport program rates. Despite thorough modeling and simulation, initial outcomes indicated that the models struggled to learn effectively, attributed potentially to oversimplified environmental assumptions. This paper discusses the challenges encountered, evaluates the models' performance against actual operational data, and outlines future directions to refine RL applications in ATM.

Fan Hu, Zijie Xin, Xirong Li

Ad-hoc Video Search (AVS) involves using a textual query to search for multiple relevant videos in a large collection of unlabeled short videos. The main challenge of AVS is the visual diversity of relevant videos. A simple query such as "Find shots of a man and a woman dancing together indoors" can span a multitude of environments, from brightly lit halls and shadowy bars to dance scenes in black-and-white animations. It is therefore essential to retrieve relevant videos as comprehensively as possible. Current solutions for the AVS task primarily fuse multiple features into one or more common spaces, yet overlook the need for diverse spaces. To fully exploit the expressive capability of individual features, we propose LPD, short for Learning Partially Decorrelated common spaces. LPD incorporates two key innovations: feature-specific common space construction and the de-correlation loss. Specifically, LPD learns a separate common space for each video and text feature, and employs de-correlation loss to diversify the ordering of negative samples across different spaces. To enhance the consistency of multi-space convergence, we designed an entropy-based fair multi-space triplet ranking loss. Extensive experiments on the TRECVID AVS benchmarks (2016-2023) justify the effectiveness of LPD. Moreover, diversity visualizations of LPD's spaces highlight its ability to enhance result diversity.

Fan Hu, Aozhu Chen, Xirong Li

While deep learning based image retrieval is reported to be vulnerable to adversarial attacks, existing works are mainly on image-to-image retrieval with their attacks performed at the front end via query modification. By contrast, we present in this paper the first study about a threat that occurs at the back end of a text-to-image retrieval (T2IR) system. Our study is motivated by the fact that the image collection indexed by the system will be regularly updated due to the arrival of new images from various sources such as web crawlers and advertisers. With malicious images indexed, it is possible for an attacker to indirectly interfere with the retrieval process, letting users see certain images that are completely irrelevant w.r.t. their queries. We put this thought into practice by proposing a novel Trojan-horse attack (THA). In particular, we construct a set of Trojan-horse images by first embedding word-specific adversarial information into a QR code and then putting the code on benign advertising images. A proof-of-concept evaluation, conducted on two popular T2IR datasets (Flickr30k and MS-COCO), shows the effectiveness of the proposed THA in a white-box mode.

Fan Hu, Aozhu Chen, Ziyue Wang et al.

In this paper we revisit feature fusion, an old-fashioned topic, in the new context of text-to-video retrieval. Different from previous research that considers feature fusion only at one end, let it be video or text, we aim for feature fusion for both ends within a unified framework. We hypothesize that optimizing the convex combination of the features is preferred to modeling their correlations by computationally heavy multi-head self attention. We propose Lightweight Attentional Feature Fusion (LAFF). LAFF performs feature fusion at both early and late stages and at both video and text ends, making it a powerful method for exploiting diverse (off-the-shelf) features. The interpretability of LAFF can be used for feature selection. Extensive experiments on five public benchmark sets (MSR-VTT, MSVD, TGIF, VATEX and TRECVID AVS 2016-2020) justify LAFF as a new baseline for text-to-video retrieval.

Aozhu Chen, Fan Hu, Zihan Wang et al.

For quantifying progress in Ad-hoc Video Search (AVS), the annual TRECVID AVS task is an important international evaluation. Solutions submitted by the task participants vary in terms of their choices of cross-modal matching models, visual features and training data. As such, what one may conclude from the evaluation is at a high level that is insufficient to reveal the influence of the individual components. In order to bridge the gap between the current solution-level comparison and the desired component-wise comparison, we propose in this paper a large-scale and systematic evaluation on TRECVID. By selected combinations of state-of-the-art matching models, visual features and (pre-)training data, we construct a set of 25 different solutions and evaluate them on the TRECVID AVS tasks 2016--2020. The presented evaluation helps answer the key question of what matters for AVS. The resultant observations and learned lessons are also instructive for developing novel AVS solutions.

Fan Hu, Lei Wang, Yishen Hu et al.

The identification of protein-ligand interaction plays a key role in biochemical research and drug discovery. Although deep learning has recently shown great promise in discovering new drugs, there remains a gap between deep learning-based and experimental approaches. Here we propose a novel framework, named AIMEE, integrating AI Model and Enzymology Experiments, to identify inhibitors against 3CL protease of SARS-CoV-2, which has taken a significant toll on people across the globe. From a bioactive chemical library, we have conducted two rounds of experiments and identified six novel inhibitors with a hit rate of 29.41%, and four of them showed an IC50 value less than 3 μM. Moreover, we explored the interpretability of the central model in AIMEE, mapping the deep learning extracted features to domain knowledge of chemical properties. Based on this knowledge, a commercially available compound was selected and proven to be an activity-based probe of 3CLpro. This work highlights the great potential of combining deep learning models and biochemical experiments for intelligent iteration and expanding the boundaries of drug discovery.

Bhavya Karki, Fan Hu, Nithin Haridas et al.

In this paper, we describe a dataset and baseline result for a question answering that utilizes web tables. It contains commonly asked questions on the web and their corresponding answers found in tables on websites. Our dataset is novel in that every question is paired with a table of a different signature. In particular, the dataset contains two classes of tables: entity-instance tables and the key-value tables. Each QA instance comprises a table of either kind, a natural language question, and a corresponding structured SQL query. We build our model by dividing question answering into several tasks, including table retrieval and question element classification, and conduct experiments to measure the performance of each task. We extract various features specific to each task and compose a full pipeline which constructs the SQL query from its parts. Our work provides qualitative results and error analysis for each task, and identifies in detail the reasoning required to generate SQL expressions from natural language questions. This analysis of reasoning informs future models based on neural machine learning.

Jin Huang, Gui-Song Xia, Fan Hu et al.

This paper aims to address the problem of detecting buildings from remote sensing images with very high resolution (VHR). Inspired by the observation that buildings are always more distinguishable in geometries than in texture or spectral, we propose a new geometric building index (GBI) for accurate building detection, which relies on the geometric saliency of building structures. The geometric saliency of buildings is derived from a mid-level geometric representations based on meaningful junctions that can locally describe anisotropic geometrical structures of images. The resulting GBI is measured by integrating the derived geometric saliency of buildings. Experiments on three public datasets demonstrate that the proposed GBI achieves very promising performance, and meanwhile shows impressive generalization capability.

Fan Hu, Gui-Song Xia, Wen Yang et al.

Scene classification is a fundamental task in interpretation of remote sensing images, and has become an active research topic in remote sensing community due to its important role in a wide range of applications. Over the past years, tremendous efforts have been made for developing powerful approaches for scene classification of remote sensing images, evolving from the traditional bag-of-visual-words model to the new generation deep convolutional neural networks (CNNs). The deep CNN based methods have exhibited remarkable breakthrough on performance, dramatically outperforming previous methods which strongly rely on hand-crafted features. However, performance with deep CNNs has gradually plateaued on existing public scene datasets, due to the notable drawbacks of these datasets, such as the small scale and low-diversity of training samples. Therefore, to promote the development of new methods and move the scene classification task a step further, we deeply discuss the existing problems in scene classification task, and accordingly present three open directions. We believe these potential directions will be instructive for the researchers in this field.

Pu Jin, Gui-Song Xia, Fan Hu et al.

Aerial image scene classification is a fundamental problem for understanding high-resolution remote sensing images and has become an active research task in the field of remote sensing due to its important role in a wide range of applications. However, the limitations of existing datasets for scene classification, such as the small scale and low-diversity, severely hamper the potential usage of the new generation deep convolutional neural networks (CNNs). Although huge efforts have been made in building large-scale datasets very recently, e.g., the Aerial Image Dataset (AID) which contains 10,000 image samples, they are still far from sufficient to fully train a high-capacity deep CNN model. To this end, we present a larger-scale dataset in this paper, named as AID++, for aerial scene classification based on the AID dataset. The proposed AID++ consists of more than 400,000 image samples that are semi-automatically annotated by using the existing the geographical data. We evaluate several prevalent CNN models on the proposed dataset, and the results show that our dataset can be used as a promising benchmark for scene classification.

Xin-Yi Tong, Gui-Song Xia, Fan Hu et al.

Remote sensing (RS) image retrieval is of great significant for geological information mining. Over the past two decades, a large amount of research on this task has been carried out, which mainly focuses on the following three core issues: feature extraction, similarity metric and relevance feedback. Due to the complexity and multiformity of ground objects in high-resolution remote sensing (HRRS) images, there is still room for improvement in the current retrieval approaches. In this paper, we analyze the three core issues of RS image retrieval and provide a comprehensive review on existing methods. Furthermore, for the goal to advance the state-of-the-art in HRRS image retrieval, we focus on the feature extraction issue and delve how to use powerful deep representations to address this task. We conduct systematic investigation on evaluating correlative factors that may affect the performance of deep features. By optimizing each factor, we acquire remarkable retrieval results on publicly available HRRS datasets. Finally, we explain the experimental phenomenon in detail and draw conclusions according to our analysis. Our work can serve as a guiding role for the research of content-based RS image retrieval.

Gui-Song Xia, Jingwen Hu, Fan Hu et al.

Aerial scene classification, which aims to automatically label an aerial image with a specific semantic category, is a fundamental problem for understanding high-resolution remote sensing imagery. In recent years, it has become an active task in remote sensing area and numerous algorithms have been proposed for this task, including many machine learning and data-driven approaches. However, the existing datasets for aerial scene classification like UC-Merced dataset and WHU-RS19 are with relatively small sizes, and the results on them are already saturated. This largely limits the development of scene classification algorithms. This paper describes the Aerial Image Dataset (AID): a large-scale dataset for aerial scene classification. The goal of AID is to advance the state-of-the-arts in scene classification of remote sensing images. For creating AID, we collect and annotate more than ten thousands aerial scene images. In addition, a comprehensive review of the existing aerial scene classification techniques as well as recent widely-used deep learning methods is given. Finally, we provide a performance analysis of typical aerial scene classification and deep learning approaches on AID, which can be served as the baseline results on this benchmark.

Jingwen Hu, Gui-Song Xia, Fan Hu et al.

Scene classification is a key problem in the interpretation of high-resolution remote sensing imagery. Many state-of-the-art methods, e.g. bag-of-visual-words model and its variants, the topic models as well as deep learning-based approaches, share similar procedures: patch sampling, feature description/learning and classification. Patch sampling is the first and a key procedure which has a great influence on the results. In the literature, many different sampling strategies have been used, {e.g. dense sampling, random sampling, keypoint-based sampling and saliency-based sampling, etc. However, it is still not clear which sampling strategy is suitable for the scene classification of high-resolution remote sensing images. In this paper, we comparatively study the effects of different sampling strategies under the scenario of scene classification of high-resolution remote sensing images. We divide the existing sampling methods into two types: dense sampling and sparse sampling, the later of which includes random sampling, keypoint-based sampling and various saliency-based sampling proposed recently. In order to compare their performances, we rely on a standard bag-of-visual-words model to construct our testing scheme, owing to their simplicity, robustness and efficiency. The experimental results on two commonly used datasets show that dense sampling has the best performance among all the strategies but with high spatial and computational complexity, random sampling gives better or comparable results than other sparse sampling methods, like the sophisticated multi-scale key-point operators and the saliency-based methods which are intensively studied and commonly used recently.