Modern Internet of Things (IoT) systems generate massive, heterogeneous multivariate time series data. Accurate Multivariate Time Series Forecasting (MTSF) of such data is critical for numerous applications. However, existing methods almost universally employ a shared embedding layer that processes all channels identically, creating a representational bottleneck that obscures valuable channel-specific information. To address this challenge, we introduce a Shared-Auxiliary Embedding (SAE) framework that decomposes the embedding into a shared base component capturing common patterns and channel-specific auxiliary components modeling unique deviations. Within this decomposition, we empirically observe that the auxiliary components tend to exhibit low-rank and clustering characteristics, a structural pattern that is significantly less apparent when using purely independent embeddings. Consequently, we design LightSAE, a parameter-efficient embedding module that operationalizes these observed characteristics through low-rank factorization and a shared, gated component pool. Extensive experiments across 9 IoT-related datasets and 4 backbone architectures demonstrate LightSAE’s effectiveness, achieving MSE improvements of up to 22.8% with only 4.0% parameter increase. Code is available at https://github.com/EDM314/LightSAE.

Pre-trained on extensive text and image corpora, current Multi-Modal Large Language Models (MLLM) have shown strong capabilities in general visual reasoning tasks. However, their performance is still lacking in physical domains that require understanding diagrams with complex physical structures and quantitative analysis based on multi-modal information. To address this, we develop a new framework, named **M**ulti-Modal Scientific Re**A**soning with **P**hysics Perception and **S**imulation (**MAPS**) based on an MLLM. MAPS decomposes expert-level multi-modal reasoning task into physical diagram understanding via a Physical Perception Model (PPM) and reasoning with physical knowledge via a simulator. The PPM module is obtained by fine-tuning a visual language model using carefully designed synthetic data with paired physical diagrams and corresponding simulation language descriptions. At the inference stage, MAPS integrates the simulation language description of the input diagram provided by PPM and results obtained through a Chain-of-Simulation process with MLLM to derive the underlying rationale and the final answer. Validated using our collected college-level circuit analysis problems, MAPS significantly improves reasoning accuracy of MLLM and outperforms all existing models. The results confirm MAPS offers a promising direction for enhancing multi-modal scientific reasoning ability of MLLMs. We will release our code, model and dataset used for our experiments upon publishing of this paper.

This paper proposed an online and offline deeply blended teaching and learning mode in the course "Principles of Electric Circuits" in Tsinghua University since 2013. Students are required to preview all contents by watching videos in MOOC, profound discussion will be hold in class time, including questions met in preview, comprehensive problems, extensions of principles application and portable experiments. Firstly, general pedagogical design of this innovative mode was introduced in detail, and the lectures of "sinusoidal steady state analysis" set as an example. Secondly, the assessment policy of the course was designed including both process and summative evaluation, which aims to stimulate the continuous effort of the student and evaluate their all-sided growth in the course. Then verified course improvement compared with parallel classes of the same course was introduced, including examination grades, questionnaire feedback and following performance of the students. Finally, several key points and our corresponding measures which are critical for the success of this teaching and learning mode were explained.

At the beginning of 2020, coronavirus disease 2019 (COVID-19) spread quickly throughout China. For the health and safety of faculties and students, all schools and universities were required to postpone their spring semester; however, the Ministry of Education hoped that teaching and learning could be facilitated to decrease or even eliminate the negative impact of the postponement on the students' subsequent self-development. Taking the "Principles of Electric Circuits" course as an example, this paper presents an innovative online teaching scheme for fundamental engineering courses, including pedagogical design and interactive methods in lectures. The students' feedback was collected using delicately designed questionnaires and the results demonstrated that our online teaching scheme achieved an effect no less than that from traditional in-classroom teaching, with even better student interactions using the online networks.

The meat grinder with CPFU (Capacitive Pulse Forming Unit) circuit is a latest pulse-compression circuit for pulsed power applications. However, one imperfection of this circuit is that it does not solve the residual energy recovery problem, i.e., to recover and reuse the residual energy after one operation. This paper presents a theoretically feasible solution by making a small but effective modification to the meat grinder with CPFU circuit. The working principle of the new circuit is illustrated stage by stage. According to numerical simulations, the overall energy recovery efficiency can reach 87% in theory.

Summary Hybrid storage systems that consist of flash‐based solid state drives (SSDs) and traditional disks are now widely used. In hybrid storage systems, there exists a two‐level cache hierarchy that regard dynamic random access memory (DRAM) as the first level cache and SSD as the second level cache for disk storage. However, this two‐level cache hierarchy typically uses independent cache replacement policies for each level, which makes cache resource management inefficient and reduces system performance. In this paper, we propose a novel adaptive multi‐level cache (AMC) replacement algorithm in hybrid storage systems. The AMC algorithm adaptively adjusts cache blocks between DRAM and SSD cache levels using an integrated solution. AMC uses combined selective promote and demote operations to dynamically determine the level in which the blocks are to be cached. In this manner, the AMC algorithm achieves multi‐level cache exclusiveness and makes cache resource management more efficient. By using real‐life storage traces, our evaluation shows the proposed algorithm improves hybrid multi‐level cache performance and also increases the SSD lifetime compared with traditional multi‐level cache replacement algorithms. Copyright © 2015 John Wiley & Sons, Ltd.