Many real-world datasets are time series that are sequentially collected and contain rich temporal information. Thus, a common interest in practice is to capture dynamics of time series and predict their future evolutions. To this end, the recurrent neural network (RNN) has been a prevalent and effective machine learning option, which admits a nonlinear state-space model representation. Motivated by the resemblance between RNN and Kalman filter (KF) for linear state-space models, we propose in this paper Innovation-driven RNN (IRNN), a novel RNN architecture tailored to time-series data modeling and prediction tasks. By adapting the concept of "innovation" from KF to RNN, past prediction errors are adopted as additional input signals to update hidden states of RNN and boost prediction performance. Since innovation data depend on network parameters, existing training algorithms for RNN do not apply to IRNN straightforwardly. Thus, a tailored training algorithm dubbed input updating-based back-propagation through time (IU-BPTT) is further proposed, which alternates between updating innovations and optimizing network parameters via gradient descent. Experiments on real-world benchmark datasets show that the integration of innovations into various forms of RNN leads to remarkably improved prediction accuracy of IRNN without increasing the training cost substantially.

The construction of experimental datasets is essential for expanding the scope of data-driven scientific discovery. Recent adva nces in natural language pro cessing (NLP) have facilitated automatic extraction of structured data from uns tructured scientific literature. While existing approaches--multi-step and direct methods--offer va luable capabilities, they also come with limitations when applied independently. He re, we propose a novel hybrid text-mining framework that integrates the advantages of both methods to convert unstructured scientific text into structured data. Our approach first tran sforms raw text into entity-recognized text, and subsequently into structured form. Furthermore, beyond the overall data structuring framework, we also enhance entity recogniti on performance by introducing an entity marker--a simple yet effective technique that uses sym bolic annotations to highlight target entities. Specifically, our entity marker-based hybrid approach not onl y consistently outperforms previous entity recognition approaches across three benchmark datasets (MatScholar, SOFC, and SOFC slot NER) but also improve the quality of final st ructured data--yielding up to a 58% improvement in entity-level F1 score and up to 83% improveme nt in relation-level F1 score compared to direct approach.

The state of health (SOH) of lithium-ion batteries (LIBs) is crucial for ensuring the safe and reliable operation of electric vehicles. Nevertheless, the prevailing SOH estimation methods often have limited generalizability. This paper introduces a data-driven approach for estimating the SOH of LIBs, which is designed to improve generalization. We construct a hybrid model named ACLA, which integrates the attention mechanism, convolutional neural network (CNN), and long short-term memory network (LSTM) into the augmented neural ordinary differential equation (ANODE) framework. This model employs normalized charging time corresponding to specific voltages in the constant current charging phase as input and outputs the SOH as well as remaining useful of life. The model is trained on NASA and Oxford datasets and validated on the TJU and HUST datasets. Compared to the benchmark models NODE and ANODE, ACLA exhibits higher accuracy with root mean square errors (RMSE) for SOH estimation as low as 1.01% and 2.24% on the TJU and HUST datasets, respectively.

Transformer-based models are the foundation of modern machine learning, but their execution, particularly during autoregressive decoding in large language models (LLMs), places significant pressure on memory systems due to frequent memory accesses and growing key-value (KV) caches. This creates a bottleneck in memory bandwidth, especially as context lengths increase. Processing-in-memory (PIM) architectures are a promising solution, offering high internal bandwidth and compute parallelism near memory. However, current PIM designs are primarily optimized for dense attention and struggle with the dynamic, irregular access patterns introduced by modern KV cache sparsity techniques. Consequently, they suffer from workload imbalance, reducing throughput and resource utilization. In this work, we propose STARC, a novel sparsity-optimized data mapping scheme tailored specifically for efficient LLM decoding on PIM architectures. STARC clusters KV pairs by semantic similarity and maps them to contiguous memory regions aligned with PIM bank structures. During decoding, queries retrieve relevant tokens at cluster granularity by matching against precomputed centroids, enabling selective attention and parallel processing without frequent reclustering or data movement overhead. Experiments on the HBM-PIM system show that, compared to common token-wise sparsity methods, STARC reduces attention-layer latency by 19%--31% and energy consumption by 19%--27%. Under a KV cache budget of 1024, it achieves up to 54%--74% latency reduction and 45%--67% energy reduction compared to full KV cache retrieval. Meanwhile, STARC maintains model accuracy comparable to state-of-the-art sparse attention methods, demonstrating its effectiveness in enabling efficient and hardware-friendly long-context LLM inference on PIM architectures.

-- Miniaturizing legged robot platforms is challenging due to hardware limitations that constrain the number, power density, and precision of actuators at that size. By leveraging design principles of quasi-passive walking robots at any scale, stable locomotion and steering can be achieved with simple mechanisms and open-loop control. Here, we present the design and control of "Zippy", the smallest self-contained bipedal walking robot at only 3 . Zippy has rounded feet, a single motor without feedback control, and is capable of turning, skipping, and ascending steps. At its fastest pace, the robot achieves a forward walking speed of 25 cm/ s, which is 10 leg lengths per second, the fastest biped robot of any size by that metric. This work explores the design and performance of the robot and compares it to similar dynamic walking robots at larger scales. Small, centimeter-scale robots have the potential to excel at traversing tight spaces found in industrial facilities, natural cavities, and disaster debris, allowing for inspection and exploration tasks typically inaccessible to robots with larger footprints.

--Recently, spiking neural networks (SNNs), deployed on neuromorphic chips, provide highly efficient solutions on edge devices in different scenarios. However, their ability to adapt to distribution shifts after deployment has become a crucial challenge. Online test-time adaptation (OTT A) offers a promising solution by enabling models to dynamically adjust to new data distributions without requiring source data or labeled target samples. Nevertheless, existing OTT A methods are largely designed for traditional artificial neural networks and are not well-suited for SNNs. T o address this gap, we propose a low-power, neuromorphic chip-friendly online test-time adaptation framework, aiming to enhance model generalization under distribution shifts. The proposed approach is called Threshold Modulation (TM), which dynamically adjusts the firing threshold through neuronal dynamics-inspired normalization, being more compatible with neuromorphic hardware. Experimental results on benchmark datasets demonstrate the effectiveness of this method in improving the robustness of SNNs against distribution shifts while maintaining low computational cost. The proposed method offers a practical solution for online test-time adaptation of SNNs, providing inspiration for the design of future neuromorphic chips. In recent years, with the rapid development of high-performance hardware and training algorithms, modern deep artifical neural networks (ANNs) can have billions, or even hundreds of billions, of parameters, requiring large-scale computational resource for training and inference.

This study aims to analyze the economic performance of various parks under different conditions, particularly focusing on the operational costs and power load balancing before and after the deployment of energy storage systems. Firstly, the economic performance of the parks without energy storage was analyzed using a random forest model. Taking Park A as an example, it was found that the cost had the greatest correlation with electricity purchase, followed by photovoltaic output, indicating that solar and wind power output are key factors affecting economic performance. Subsequently, the operation of the parks after the configuration of a 50kW/100kWh energy storage system was simulated, and the total cost and operation strategy of the energy storage system were calculated. The results showed that after the deployment of energy storage, the amount of wind and solar power curtailment in each park decreased, and the operational costs were reduced. Finally, a genetic algorithm was used to optimize the energy storage configuration of each park. The energy storage operation strategy was optimized through fitness functions, crossover operations, and mutation operations. After optimization, the economic indicators of Parks A, B, and C all improved. The research results indicate that by optimizing energy storage configuration, each park can reduce costs, enhance economic benefits, and achieve sustainable development of the power system.

Air pollution is a significant global health risk, contributing to millions of premature deaths annually. Nitrogen dioxide (NO2), a harmful pollutant, disproportionately affects urban areas where monitoring networks are often sparse. We propose a novel method for predicting NO2 concentrations at unmonitored locations using transfer learning with satellite and meteorological data. Leveraging the GraphSAGE framework, our approach integrates autoregression and transfer learning to enhance predictive accuracy in data-scarce regions like Bristol. Pre-trained on data from London, UK, our model achieves a 8.6% reduction in Normalised Root Mean Squared Error (NRMSE) and a 32.6% reduction in Gradient RMSE compared to a baseline model. This work demonstrates the potential of virtual sensors for cost-effective air quality monitoring, contributing to actionable insights for climate and health interventions.

Sen. John Fetterman, D-Pa., receives praise for his less-than-formal attire from Sam Altman during a Commerce Committee hearing. Sen. John Fetterman, D-Pa., was one of the final senators to question OpenAI chief Sam Altman during Thursday's Senate Commerce Committee hearing, and the subject of both Three Mile Island and the Democrat's penchant for Carhartt outerwear came up. Fetterman said that as a senator he has been able to meet people with "much more impressive jobs and careers" and that due to Altman's technology, "humans will have a wonderful ability to adapt." He told Altman that some Americans are worried about AI on various levels, and he asked the executive to address it. In response, Altman said he appreciated Fetterman's praise.

The recent progress of using graph based encoding of crystal structures for high throughput material property prediction has been quite successful. However, using a single modality model prevents us from exploiting the advantages of an enhanced features space by combining different representations. Specifically, pre-trained Large language models(LLMs) can encode a large amount of knowledge which is beneficial for training of models. Moreover, the graph encoder is able to learn the local features while the text encoder is able to learn global information such as space group and crystal symmetry. In this work, we propose Material Multi-Modal Fusion(MatMMFuse), a fusion based model which uses a multi-head attention mechanism for the combination of structure aware embedding from the Crystal Graph Convolution Network (CGCNN) and text embeddings from the SciBERT model. We train our model in an end-to-end framework using data from the Materials Project Dataset. We show that our proposed model shows an improvement compared to the vanilla CGCNN and SciBERT model for all four key properties: formation energy, band gap, energy above hull and fermi energy. Specifically, we observe an improvement of 40% compared to the vanilla CGCNN model and 68% compared to the SciBERT model for predicting the formation energy per atom. Importantly, we demonstrate the zero shot performance of the trained model on small curated datasets of Perovskites, Chalcogenides and the Jarvis Dataset. The results show that the proposed model exhibits better zero shot performance than the individual plain vanilla CGCNN and SciBERT model. This enables researchers to deploy the model for specialized industrial applications where collection of training data is prohibitively expensive.