Human identification at a distance (HID) faces challenges due to the difficulty of acquiring traditional biometric modalities like face and fingerprints. Gait recognition offers a viable solution since it can be captured at a distance. To promote progress in gait recognition and provide a fair evaluation platform, the International Competition on Human Identification at a Distance (HID) has been organized annually since 2020. Since 2023, the competition has adopted the challenging SUSTech-Competition dataset, which includes significant variations in clothing, carried objects, and view angles. No training data is provided, requiring participants to train their models using external datasets. Each year, the competition applies a different random seed to generate distinct evaluation splits, reducing the risk of overfitting and ensuring fair evaluation of cross-domain generalization. Although the previous two competitions (HID 2023 and HID 2024) already utilized this dataset, HID 2025 aimed explicitly to explore whether algorithmic improvements could surpass the accuracy limits observed previously. Despite these heightened challenges, participants again demonstrated significant advancements, with the highest accuracy reaching 94.2%, setting a new benchmark for this dataset. We also analyze key technical trends and outline potential directions for future research on gait recognition.

The Multimodal Large Language Model (MLLM) is currently experiencing rapid growth, driven by the advanced capabilities of language-based LLMs. Unlike their specialist predecessors, existing MLLMs are evolving towards a Multimodal Generalist paradigm. Initially limited to understanding multiple modalities, these models have advanced to not only comprehend but also generate across modalities. Their capabilities have expanded from coarse-grained to fine-grained multimodal understanding and from supporting singular modalities to accommodating a wide array of or even arbitrary modalities. To assess the capabilities of various MLLMs, a diverse array of benchmark test sets has been proposed. This leads to a critical question: Can we simply assume that higher performance across tasks indicates a stronger MLLM capability, bringing us closer to human-level AI? We argue that the answer is not as straightforward as it seems. In this project, we introduce an evaluation framework to delineate the capabilities and behaviors of current multimodal generalists. This framework, named General-Level, establishes 5-scale levels of MLLM performance and generality, offering a methodology to compare MLLMs and gauge the progress of existing systems towards more robust multimodal generalists and, ultimately, towards AGI (Artificial General Intelligence). Central to our framework is the use of Synergy as the evaluative criterion, categorizing capabilities based on whether MLLMs preserve synergy across comprehension and generation, as well as across multimodal interactions. To evaluate the comprehensive abilities of various generalists, we present a massive multimodal benchmark, General-Bench, which encompasses a broader spectrum of skills, modalities, formats, and capabilities, including over 700 tasks and 325,800 instances. The evaluation results that involve over 100 existing state-of-the-art MLLMs uncover the capability rankings of generalists, highlighting the challenges in reaching genuine AI. We expect this project to pave the way for future research on next-generation multimodal foundation models, providing a robust infrastructure to accelerate the realization of AGI. Project Page: https://generalist.top/, Leaderboard: https://generalist.top/leaderboard/, Benchmark: https://huggingface.co/General-Level/.

This article is concerned with the rapid classification issue for dynamical patterns consisting of sampling sequences in a relatively large-scale dynamical dataset constructed by benchmark Rossler systems. Specifically, based on a recently developed deterministic learning mechanism, a rapid dynamical pattern classification method is developed, which contains a modeling stage and a classification stage. In the modeling stage, a deterministic learning scheme is employed to accurately learn/model the inherent dynamics of the training dynamical patterns and store the acquired knowledge in a set of constant radial basis function (RBF) networks. In the classification stage, based on the trained RBF networks, a set of dynamical estimators is developed for real-time dynamic comparison. The generating recognition errors are then used to effectively represent the dynamic differences in real-time. To this end, the associated class label of the minimum recognition error is assigned to the test pattern also in real-time. To demonstrate the effectiveness of the proposed method, a relatively large-scale dynamical pattern dataset containing various dynamical behaviors is constructed by utilizing a deterministic chaos prospector (DCP) technique. The simulation results show that the new method achieves competitive classification performances compared to the state-of-the-art time-series classification method for the dynamical system classification task. In addition to performance advantages, the new method can perform real-time time-series classification with the first 10% of data achieving over 95% of accuracy based on the full-length data. Besides, the superiority of our method is demonstrated from various datasets in the UCR time-series classification (TSC) archive.

Human identification at a distance (HID) faces challenges due to the difficulty of acquiring traditional biometric modalities like face and fingerprints. Gait recognition offers a viable solution since it can be captured at a distance. To advance the algorithm development and provide fair evaluations, the International Competition on Human Identification at a Distance (HID) has been held annually since 2020, with HID 2024 marking the fifth edition. Despite increased difficulty, participants demonstrated remarkable capabilities, surpassing previous accuracy levels. This paper, co-authored by competition organizers and top participants, provides a comprehensive summary of HID 2024, including an overview of the competition, and insights into the methods employed by the top teams. Specifically, inspired by the achievements of the 5 competitions of HID, we also provide the insights for the future directions on gait recognition.

A key challenge for training control policies with imitation learning methods lies in the computational inefficiency. This inefficiency comes from an assumption that underlies these methods, assuming the policy should compute a new action for each state, which is unnecessary and costly. However, we notice the states occurring within K consecutive timesteps differ negligibly and their corresponding actions are extremely similar. Therefore, we challenge this assumption and argue that it is enough to compute an action every K states. With this argument, we propose K-Timestep Adaptive Action Approximation, which replaces the computation of K one-timestep actions approximately with that of one K-timestep action to alleviate the computational inefficiency issue. To demonstrate the theoretical validity of our method, we analyze the errors incurred by the policies learned via the method. The analysis proves these policies can converge to the optimal solution with errors no more than an upper bound dependent on K, revealing the effectiveness of our method. To avoid the difficulty of hyperparameter handcrafting on K, we design a simple but effective auto-hyperparameter tuning strategy. In the proposed strategy, K is added as an extra dimension to the action space of the policy, so it can be tuned adaptively by the policy according to the state without any user intervention. Empirical results on 4 imitation learning tasks show the superiority in computational efficiency of our method, which can effectively reduce new actions to compute in training policies.

Rapid dynamical pattern recognition based on the deterministic learning method (DLM-based RDPR) aims to rapidly recognize the most similar dynamical pattern pair from perspectives of differences in inherent system dynamics. The basic mechanism is to use available recognition errors to reflect the differences in the dynamics of dynamical pattern pairs and then to make a decision based on a minimal recognition error (MRE) principle. This article focuses on providing a rigorous theoretical analysis of the MRE principle in DLM-based RDPR under the sampled-data framework. Specifically, we seek a unified methodology from the similarity definition to the measure implementation and then to derive general sufficient conditions and necessary conditions for the MRE principle. The main idea is to: 1) from the average signal energy aspect, define a time-dependent dynamics-based similarity in dynamical pattern pairs and reestablish the measure of recognition errors generated from the DLM-based RDPR; 2) introduce the energy-based Lyapunov method to establish the interrelation between the dynamical distance and the recognition error; and 3) derive sufficient conditions and necessary conditions from two directions of the interrelation. The proposed conditions distinguish themselves from virtually all of the existing DLM-based RDPR works with only sufficient conditions in the sense that it is shown in a rigorous analysis that under what conditions, the pattern pair recognized based on the MRE principle is indeed the most similar one. Therefore, the proposed work makes the DLM-based RDPR possess good interpretability and provides strong theoretical guidance in engineering applications.

This article investigates dynamical pattern recognition for a class of univariate time-series data. These data are sampled from the output of dynamical systems with uncertain dynamics. Based on deterministic learning, a rapid recognition approach is presented from the viewpoint of the sample-data observer. It comprises two phases: 1) training and 2) recognition. In the training phase, locally accurate dynamical modeling of the underlying dynamics of training time series can be accomplished by merging a sampled-data observer and radial basis function network (RBFN) identifiers. In the recognition phase, several RBFN-based estimators with non-high-gain designs are constructed. In this case, the stability analysis of the generated estimator error systems will conduce to conduct non-high-gain recognition of a test time series. We demonstrate that these estimator errors can depict dynamics differences between the dynamical patterns of the test and training time-series data. Based on the average$L_{1}$norms of the output errors, a decision-making scheme is developed to generate recognition results rapidly. More concise and relaxed recognition conditions are derived through rigorous analysis to ensure accurate recognition results. Simulation studies exemplify the effectiveness of the presented approach.

BACKGROUND: Osteoporosis is a common metabolic skeletal disease and usually lacks obvious symptoms. Many individuals are not diagnosed until osteoporotic fractures occur. Bone mineral density (BMD) measured by dual-energy X-ray absorptiometry (DXA) is the gold standard for osteoporosis detection. However, only a limited percentage of people with osteoporosis risks undergo the DXA test. As a result, it is vital to develop methods to identify individuals at-risk based on methods other than DXA. RESULTS: We proposed a hierarchical model with three layers to detect osteoporosis using clinical data (including demographic characteristics and routine laboratory tests data) and CT images covering lumbar vertebral bodies rather than DXA data via machine learning. 2210 individuals over age 40 were collected retrospectively, among which 246 individuals' clinical data and CT images are both available. Irrelevant and redundant features were removed via statistical analysis. Consequently, 28 features, including 16 clinical data and 12 texture features demonstrated statistically significant differences (p < 0.05) between osteoporosis and normal groups. Six machine learning algorithms including logistic regression (LR), support vector machine with radial-basis function kernel, artificial neural network, random forests, eXtreme Gradient Boosting and Stacking that combined the above five classifiers were employed as classifiers to assess the performances of the model. Furthermore, to diminish the influence of data partitioning, the dataset was randomly split into training and test set with stratified sampling repeated five times. The results demonstrated that the hierarchical model based on LR showed better performances with an area under the receiver operating characteristic curve of 0.818, 0.838, and 0.962 for three layers, respectively in distinguishing individuals with osteoporosis and normal BMD. CONCLUSIONS: The proposed model showed great potential in opportunistic screening for osteoporosis without additional expense. It is hoped that this model could serve to detect osteoporosis as early as possible and thereby prevent serious complications of osteoporosis, such as osteoporosis fractures.

A unified approach is proposed to design sampled-data observers for a certain type of unknown nonlinear systems undergoing recurrent motions based on deterministic learning in this article. First, a discrete-time implementation of high-gain observer (HGO) is utilized to obtain state trajectory from sampled output measurements. By taking the recurrent estimated trajectory as inputs to a dynamical radial basis function network (RBFN), a partial persistent exciting (PE) condition is satisfied, and a locally accurate approximation of nonlinear dynamics can be realized along the estimated sampled-data trajectory. Second, an RBFN-based observer consisting of the obtained dynamics from the process of deterministic learning is designed. Without resorting to high gains, the RBFN-based observer is shown capable of achieving correct state observation. The novelty of this article lies in that, by incorporating deterministic learning with the discrete-time HGO, the nonlinear dynamics can be accurately approximated along the estimated trajectory, and such obtained knowledge can then be utilized to realize nonhigh-gain state estimation for the same or similar sampled-data systems. Simulation is performed to validate the effectiveness of the proposed approach.