The rise of Large Reasoning Models (LRMs) signifies a paradigm shift toward advanced computational reasoning. Yet, this progress disrupts traditional agent frameworks, traditionally anchored by execution-oriented Large Language Models (LLMs). To explore this transformation, we propose the LaRMA framework, encompassing nine tasks across Tool Usage, Plan Design, and Problem Solving, assessed with three top LLMs (e.g., Claude3.5-sonnet) and five leading LRMs (e.g., DeepSeek-R1). Our findings address four research questions: LRMs surpass LLMs in reasoning-intensive tasks like Plan Design, leveraging iterative reflection for superior outcomes; LLMs excel in execution-driven tasks such as Tool Usage, prioritizing efficiency; hybrid LLM-LRM configurations, pairing LLMs as actors with LRMs as reflectors, optimize agent performance by blending execution speed with reasoning depth; and LRMs' enhanced reasoning incurs higher computational costs, prolonged processing, and behavioral challenges, including overthinking and fact-ignoring tendencies. This study fosters deeper inquiry into LRMs' balance of deep thinking and overthinking, laying a critical foundation for future agent design advancements.

Large Language Models (LLMs) have demonstrated strong capabilities across various domains, with recent advancements in challenging reasoning tasks such as mathematics and programming. However, solving reasoning tasks often requires long decoding chains (of thoughts), which incur $O(N)$ time and memory consumption, where $N$ is the chain length. To mitigate $O(N)$ time and memory consumption, existing sparsity-based algorithms propose retaining only the most critical token's intermediate data (i.e., key-value cache) and discarding the rest. However, these existing algorithms struggle with the ``impossible trinity'' of accuracy, time, and memory. For example, the state-of-the-art algorithm, Quest, achieves high accuracy with $O(L)$ time but $O(N)$ memory ($L$ is the cache budget, $L \ll N$). To address this issue, in this paper, we identify a new attention pattern during the decode stage of reasoning tasks, where milestone tokens (analogous to lemmas in mathematical proofs) emerge, are utilized, and then become unimportant afterward. Based on this pattern, we propose a new algorithm named RaaS that identifies and retains milestone tokens only until they are no longer needed, achieving high accuracy with $O(L)$ time and $O(L)$ memory complexity.

Large Language Models (LLMs) are critical for a wide range of applications, but serving them efficiently becomes increasingly challenging as inputs become more complex. Context caching improves serving performance by exploiting inter-request dependency and reusing key-value (KV) cache across requests, thus improving time-to-first-token (TTFT). However, existing prefix-based context caching requires exact token prefix matches, limiting cache reuse in few-shot learning, multi-document QA, or retrieval-augmented generation, where prefixes may vary. In this paper, we present EPIC, an LLM serving system that introduces position-independent context caching (PIC), enabling modular KV cache reuse regardless of token chunk position (or prefix). EPIC features two key designs: AttnLink, which leverages static attention sparsity to minimize recomputation for accuracy recovery, and KVSplit, a customizable chunking method that preserves semantic coherence. Our experiments demonstrate that Epic delivers up to 8x improvements in TTFT and 7x throughput over existing systems, with negligible or no accuracy loss. By addressing the limitations of traditional caching approaches, Epic enables more scalable and efficient LLM inference.

Brain-computer interfaces (BCIs) represent a cutting-edge technological frontier, offering a transformative approach to human-computer interaction. By facilitating direct neural communication, BCIs enable individuals to control external devices or systems through cerebral activity alone, bypassing conventional motor pathways.BCIs are particularly promising for applications in healthcare, rehabilitation, entertainment, and education. In the medical field, they provide a glimmer of hope for individuals with motor impairments, enabling the restoration of control over bodily functions. For example, BCIs have been instrumental in assisting individuals with spinal cord injuries to operate prosthetic limbs and have aided stroke survivors in regaining mobility [1, 2]. A critical BCI modality is EEG-based motor imagery (MI), which utilizes electroencephalographic (EEG) signals to deduce a user's intent for limb movement.

As a revolutionary generative paradigm of deep learning, generative adversarial networks (GANs) have been widely applied in various fields to synthesize realistic data. However, it is challenging for conventional GANs to synthesize raw signal data, especially in some complex cases. In this paper, we develop a novel GAN framework for radio generation called "Radio GAN". Compared to conventional methods, it benefits from three key improvements. The first is learning based on sampling points, which aims to model an underlying sampling distribution of radio signals. The second is an unrolled generator design, combined with an estimated pure signal distribution as a prior, which can greatly reduce learning difficulty and effectively improve learning precision. Finally, we present an energy-constrained optimization algorithm to achieve better training stability and convergence. Experimental results with extensive simulations demonstrate that our proposed GAN framework can effectively learn transmitter characteristics and various channel effects, thus accurately modeling for an underlying sampling distribution to synthesize radio signals of high quality.

The computational prediction of wave propagation in dam-break floods is a long-standing problem in hydrodynamics and hydrology. Until now, conventional numerical models based on Saint-Venant equations are the dominant approaches. Here we show that a machine learning model that is well-trained on a minimal amount of data, can help predict the long-term dynamic behavior of a one-dimensional dam-break flood with satisfactory accuracy. For this purpose, we solve the Saint-Venant equations for a one-dimensional dam-break flood scenario using the Lax-Wendroff numerical scheme and train the reservoir computing echo state network (RC-ESN) with the dataset by the simulation results consisting of time-sequence flow depths. We demonstrate a good prediction ability of the RC-ESN model, which ahead predicts wave propagation behavior 286 time-steps in the dam-break flood with a root mean square error (RMSE) smaller than 0.01, outperforming the conventional long short-term memory (LSTM) model which reaches a comparable RMSE of only 81 time-steps ahead. To show the performance of the RC-ESN model, we also provide a sensitivity analysis of the prediction accuracy concerning the key parameters including training set size, reservoir size, and spectral radius. Results indicate that the RC-ESN are less dependent on the training set size, a medium reservoir size K=1200~2600 is sufficient. We confirm that the spectral radius \r{ho} shows a complex influence on the prediction accuracy and suggest a smaller spectral radius \r{ho} currently. By changing the initial flow depth of the dam break, we also obtained the conclusion that the prediction horizon of RC-ESN is larger than that of LSTM.