In this paper, we introduce analytic federated learning (AFL), a new training paradigm that brings analytical (i.e., closed- form) solutions to the federated learning (FL) with pre-trained models. Our AFL draws inspiration from analytic learning—a gradient-free technique that trains neural networks with analytical solutions in one epoch. In the local client training stage, the AFL facilitates a one-epoch training, eliminating the necessity for multi-epoch updates. In the aggregation stage, we derive an absolute aggregation (AA) law. This AA law allows a single-round aggregation, reducing heavy communication overhead and achieving fast convergence by removing the need for multiple aggregation rounds. More importantly, the AFL exhibits a property that invariance to data partitioning, meaning that regardless of how the full dataset is distributed among clients, the aggregated result remains identical. This could spawn various potentials, such as data heterogeneity invariance and client-number invariance. We conduct experiments across various FL settings including extremely non-IID ones, and scenarios with a large number of clients (e.g., ≥ 1000). In all these settings, our AFL constantly performs competitively while existing FL techniques encounter various obstacles. Our codes are available at https://github.com/ZHUANGHP/Analytic-federated-learning.

Traditional pattern recognition methods face challenges in pedestrian localization. In complex motion environments, failure to consider diversity, continuity and randomness of pedestrian gait can cause inaccurate localization. In this article, a feedback feature learning system is designed to improve human motion recognition and localization accuracy. First, advanced features are extracted using out-of-bag (OOB) evaluation, and optimal motion patterns are extracted using frequency pie probability, with an average recognition accuracy as high as 99.8%. Then, to enhance the accuracy of zero-velocity update (ZUPT) at different step frequencies, a novel peak detection method is developed to extract real-time stride frequency, and an adaptive step frequency-based ZUPT optimal threshold function is obtain-ed. Next, a motion constraint model is designed to overcome the influence of non-Gaussian noise, and the 3-D position information in different motion modes solved by Extended Kalman filter (EKF) is optimized. Finally, the obtained more accurate 3-D position information is used as the new feature input of the classifier to further improve the pattern recognition accuracy. After three feature iteration feedbacks, the localization accuracy in relatively complex random motion is 0.1% to 1.69%.

Class-incremental learning (CIL) under an exemplar-free constraint has presented a significant challenge. Existing methods adhering to this constraint are prone to catastrophic forgetting, far more so than replay-based techniques that retain access to past samples. In this paper, to solve the exemplar-free CIL problem, we propose a Dual-Stream Analytic Learning (DS-AL) approach. The DS-AL contains a main stream offering an analytical (i.e., closed-form) linear solution, and a compensation stream improving the inherent under-fitting limitation due to adopting linear mapping. The main stream redefines the CIL problem into a Concatenated Recursive Least Squares (C-RLS) task, allowing an equivalence between the CIL and its joint-learning counterpart. The compensation stream is governed by a Dual-Activation Compensation (DAC) module. This module re-activates the embedding with a different activation function from the main stream one, and seeks fitting compensation by projecting the embedding to the null space of the main stream's linear mapping. Empirical results demonstrate that the DS-AL, despite being an exemplar-free technique, delivers performance comparable with or better than that of replay-based methods across various datasets, including CIFAR-100, ImageNet-100 and ImageNet-Full. Additionally, the C-RLS' equivalent property allows the DS-AL to execute CIL in a phase-invariant manner. This is evidenced by a never-before-seen 500-phase CIL ImageNet task, which performs on a level identical to a 5-phase one. Our codes are available at https://github.com/ZHUANGHP/Analytic-continual-learning.

Large language models (LLMs) have shown dominant performance in various language tasks, including code-writing, machine translation, and semantic comprehension. With prompt engineering, LLMs can also comprehend complex tasks and translate them into executable code. These powers offer great potential for controlling the motion of robots. In this paper, we focus on leveraging the ability of LLMs, prompt engineering, and predefined robot action APIs to facilitate high-level motion planning for quadruped robots. With LLMs, we enable the robot to autonomously plan and execute sophisticated actions based on the comprehension of effective prompts. Through various experiments and evaluations, we demonstrate the effectiveness and adaptability of our approach in handling intricate motion tasks. Our research contributes to the advancement of intelligent robotics and paves the way for more versatile quadruped robots in real-world scenarios.

Class incremental learning (CIL) trains a network on sequential tasks with separated categories in each task but suffers from catastrophic forgetting, where models quickly lose previously learned knowledge when acquiring new tasks. The generalized CIL (GCIL) aims to address the CIL problem in a more real-world scenario, where incoming data have mixed data categories and unknown sample size distribution. Existing attempts for the GCIL either have poor performance or invade data privacy by saving exemplars. In this paper, we propose a new exemplar-free GCIL technique named generalized analytic continual learning (GACL). The GACL adopts analytic learning (a gradient-free training technique) and delivers an analytical (i.e., closed-form) solution to the GCIL scenario. This solution is derived via decomposing the incoming data into exposed and unexposed classes, thereby attaining a weight-invariant property, a rare yet valuable property supporting an equivalence between incremental learning and its joint training. Such an equivalence is crucial in GCIL settings as data distributions among different tasks no longer pose challenges to adopting our GACL. Theoretically, this equivalence property is validated through matrix analysis tools. Empirically, we conduct extensive experiments where, compared with existing GCIL methods, our GACL exhibits a consistently leading performance across various datasets and GCIL settings. Source code is available at https://github.com/CHEN-YIZHU/GACL.

Online Class Incremental Learning (OCIL) aims to train models incrementally, where data arrive in mini-batches, and previous data are not accessible. A major challenge in OCIL is Catastrophic Forgetting, i.e., the loss of previously learned knowledge. Among existing baselines, replay-based methods show competitive results but requires extra memory for storing exemplars, while exemplar-free (i.e., data need not be stored for replay in production) methods are resource friendly but often lack accuracy. In this paper, we propose an exemplar-free approach—Forward-only Online Analytic Learning (F-OAL). Unlike traditional methods, F-OAL does not rely on back-propagation and is forward-only, significantly reducing memory usage and computational time. Cooperating with a pre-trained frozen encoder with Feature Fusion, F-OAL only needs to update a linear classifier by recursive least square. This approach simultaneously achieves high accuracy and low resource consumption. Extensive experiments on bench mark datasets demonstrate F-OAL’s robust performance in OCIL scenarios. Code is available at: https://github.com/liuyuchen-cz/F-OAL

In modern life, emotions affect people in all aspects of their work and life. Long-term emotional problems tend to cause physical and mental problems such as depression. The photoplethysmogram(PPG) and electroencephalogram (EEG) etc are often used for emotion recognition, because the emotional state produces reactions from different biological systems of the human body. This article uses an internationally released emotional classification database called DEAP dataset that contains 32-channel standard EEG signals and 8-channel peripheral physiological signals. To achieve wearable emotion continuous recognition, this paper strives to obtain satisfactory emotion recognition through fewer EEG channels and peripheral physiological signals. Machine learning methods are used to classify arousal and valence that are often used in emotional recognition. The results show that the arousal classification accuracy can achieve 68% and the valence classification accuracy can achieve 66% with 5 channel EEG signal and one channel PPG signal, which can be monitored through a wearable headband.

This paper presents the design, implementation and test of a watch with the capability to measure multiple physiological signals, such as photoplethysmography (PPG), electrocardiography (ECG), and some physiological parameters extracted from PPG and ECG. Based on the PPG and ECG signals, the pulse rate and heart rate could be extracted respectively. Moreover, the pulse transmit time (PTT) could be extracted combining PPG and ECG, which was the key to calculate blood pressure (BP). The circumstance conditions (temperature, humidity and barometric pressure) were detected and displayed on the screen. A motion state was used to toggle the watch screen. Meanwhile all data measured could be sent to an android smart phone, using a Texas Instruments' (TI's) Bluetooth low energy chip CC2540. The PPG wave, ECG wave, heart rate, as well as the systolic blood pressure (SBP) and diastolic blood pressure (DBP) can be displayed on the android client in real time. Compared with the sport band and other smart watches, this multi-parameter health monitoring watch had the advantage of monitoring more parameters. We had made a test to compare this watch with standard equipment, Benavidez T5 and SUNTECH OSAR2, and the result showed that the relative error of heart rate, pulse rate, SBP and DBP against reference were 1.76 ±1.64bpm, 1.76 ±1.64bpm, 4.01 ± 2.75 mmHg, 4.64 ± 3.31 mmHg. These results demonstrated that this watch has a broad prospect in the field of mobile health care.

The death of the patients is an important event in the intensive care unit (ICU), mortality risk prediction thus offers much information for clinical decision making. However, Patient ICU mortality prediction faces challenges in many aspects, such as high dimensionality, imbalance distribution. In this paper, we modified the cost-sensitive principal component analysis (CSPCA), which is denoted by MCSPCA, to solve these problems. This modified method not only reduced the feature dimensionality but also better handled the imbalanced problem of the benchmark data. A support vector machine (SVM) model was used as the classifier to identify ICU mortality risk. As for SVM parameters optimization, a chaos particle swarm optimization (CPSO) was used to optimize the penalty parameter and the kernel parameter. The proposed model was compared with several contrast models (such as model without PCA). The test results indicate that the proposed model showed highest AUC of 0.7718 and minimum consumption time of 814s.

Non-isolated Weinberg Converter is suitable for battery discharge regulator due to the advantages of continuous current, no RHP zeroes and high efficiency. The working principle under overlap and non-overlap modes with stray parameters is analyzed in this paper. In order to estimate the heat distribution, the power loss is calculated and the trend of change in efficiency is achieved. In addition, the experimental results are given to verify the advantages of Non-isolated Weinberg Converter and the effectiveness of power loss analysis by a 600W prototype.

The development of shore-to-ship power supply technology can effectively reduce port pollution and promote the construction of green port. This paper briefly introduces the technical scheme of the common used inverter for shore power, and then puts forward the technical scheme of multi-terminal medium-voltage DC (MVDC), according to the transmission characteristics of the shore-to-ship power supply. Considering the different demand of converters connected to the distribution network and the ship grid, the paper discusses the suitable converter schemes. Based on the current operation demand of the distribution network and the new energy development trend of ports and ships in future, advantages and disadvantages of the inverter technology and the multi-terminal MVDC technology are compared, which demonstrates the feasibility and superiority of the multi-terminal MVDC technology in shore-to-ship power supply applications.