To accelerate Mixture-of-Experts (MoE) inference, the hybrid parallelism paradigm is first applying pipeline parallelism (PP) to vertically divide the model into stages, with each stage further divided horizontally using tensor or expert parallelism. On the algorithm side, dynamic Top-K routing reduces computation by activating fewer experts per token on average. In this paper, we explore the application of dynamic Top-K routing to PP-enabled MoE inference, aiming to fully unleash their combined potential. We identify key performance bottlenecks arising from Top-K value variation across layers, which conflicts with PP's typically uniform stage partitioning, as well as opportunities to optimize memory usage through their integration. To address these challenges, we present SMIDT, an efficient MoE inference framework tailored for dynamic Top-K routing. SMIDT features: (1) an adaptive, module-level uneven partitioning strategy to balance computation across PP stages, (2) a memory-aware expert replication scheme (DPMoE) that reduces communication overhead, and (3) a lightweight search algorithm combining binary search and dynamic programming to generate efficient parallelism plans. We implement SMIDT on SGLang, a state-of-the-art LLM inference framework, evaluate it on 32 A40 GPUs and 16 A100 GPUs, and compare with manually tuned parallelism strategies. Experimental results show that, when co-locating prefill and decoding phases, SMIDT achieves 1.20–3.13x throughput improvements for prefill-only tasks and 1.05–1.89x for prefill-decoding tasks. When disaggregating prefill and decoding tasks, SMIDT improves average and P99 time-to-first-token (TTFT) by 1.10–1.17x and 1.21–1.26x, respectively.

Channel knowledge map (CKM) is a promising technique that enables environment-aware wireless networks by utilizing location-specific channel prior information to improve communication and sensing performance. A fundamental problem for CKM construction is how to utilize partially observed channel knowledge data to reconstruct a complete CKM for all possible locations of interest. This problem resembles the long-standing ill-posed inverse problem, which tries to infer from a set of limited observations the cause factors that produced them. By utilizing the recent advances of solving inverse problems with generative artificial intelligence (AI), in this paper, we propose generative CKM construction method using partially observed data by solving inverse problems with diffusion models. Simulation results show that the proposed method significantly improves the performance of CKM construction compared with benchmarking schemes.

With the increasing demand for real-time channel state information (CSI) in sixth-generation (6G) mobile communication networks, channel knowledge map (CKM) emerges as a promising technique, offering a site-specific database that enables environment-awareness and significantly enhances communication and sensing performance by leveraging a priori wireless channel knowledge. However, efficient construction and utilization of CKMs require high-quality, massive, and location-specific channel knowledge data that accurately reflects the real-world environments. Inspired by the great success of ImageNet dataset in advancing computer vision and image understanding in artificial intelligence (AI) community, we introduce CKMImageNet, a dataset developed to bridge AI and environment-aware wireless communications and sensing by integrating location-specific channel knowledge data, high-fidelity environmental maps, and their visual representations. CKMImageNet supports a wide range of AI-driven approaches for CKM construction with spatially consistent and location-specific channel knowledge data, including both supervised and unsupervised, as well as discriminative and generative AI methods. The dataset is built using advanced ray-tracing techniques, ensuring high fidelity and environmental accuracy. By addressing key challenges in CKM construction and enabling AI models to learn environment-aware propagation patterns, CKMImageNet may serve as a foundational tool for advancing environment-aware 6G systems, ranging from network planning such as communication base station (BS) site selection and sensing anchor node placement, to pro-active resource allocation such as beam alignment, power allocation, interference avoidance, clutter rejection, and robot trajectory planning. Compared with existing datasets like RadioMapSeer, CKMImageNet not only provides numerical and visual representation to channel gain values, but also more diversified channel knowledge like multipath angles of arrival (AoAs) and path delays. Moreover, the dataset offers images with multiple sizes to cater to different application scenarios.

Current mainstream biometric user authentication approaches are based on passive measurements of the subject's characteristics, and usually come with less-than-satisfactory accuracy. This paper takes a unique approach to biometric authentication. Specifically, instead of training a machine learning algorithm to recognize a legitimate user, the paper proposes a hybrid type of training, in which the legitimate user is also trained to use a customized instance of the machine. The user thus achieves a level of artificially-induced expertise to interact with the machine, which makes the user easier to recognize. We implement this concept in a mouse-based user authentication system, in which we produce customized machine instances by introducing an angle offset to the standard mouse. Human subjects then rely on their kinesthetic intelligence to achieve motor learning and visual-motor adaptation to the modified mouse. We design a 7-week IRB-approved experiment, collect data from 18 human subjects over this period, and evaluate the proposed approach with two existing state-of-the-art mouse-based authentication schemes. We find that, in both schemes, our approach significantly outperforms the baseline in which a regular unaltered mouse is used. Somewhat surprisingly, results also show that our approach improves the authentication performance even when both legitimate and non-legitimate users are trained to exactly the same instance of customized machine (i.e., the same mouse angle offset). In addition, we also observe that users can generally maintain their learned expertise even after one week of washout, which further demonstrates the practicality of the approach. Finally, we present a practical strategy to manage the enrollment of users in such a proposed system.

Batteryless energy-harvesting sensor nodes can operate indefinitely, but if the harvesting rate is too low, they must operate intermittently. Intermittent operation imposes various challenges upon the system. One of the least-studied is communication–if nodes are unpowered for long, unpredictable periods of time, how can they reliably communicate with each other? In prior work, we proposed the concept of lifecycle management protocols (LMPs) to mitigate this issue and enable wireless communication directly between intermittent sensor nodes using active radios. In this paper, we propose a design framework for a class of LMPs. We then provide analytical models for the delay and throughput of two-node communication using this framework. Finally, we implement this framework on hardware and validate our models in an experimental setting. To the best of our knowledge, this is the first design framework for, and implementation of, protocols for enabling and improving general-purpose communication between intermittent sensor nodes using active radios.

Biometric user authentication is at the core of multifactor authentication, and mouse-based biometric authentication comes at no additional cost for most computer systems. This paper describes a mouse-based user authentication scheme, called MAUSPAD, which uses a novel progress-adjusted dynamic time warping (PADTW) algorithm, along with a segmentation algorithm, to accurately and meaningfully measure the differences between observed data and reference data. By introducing a new concept, which we call progress, into standard DTW, the new PADTW can have better control of the warping and mapping process and hence is more suitable for comparing time-stamped spatial sequences such as mouse cursor movements. Furthermore, in order to preserve the important but transient details in the cursor movement (which may be critical in identifying a specific user), we apply a segmentation algorithm to divide each reference cursor movement into multiple smaller segments, and measure the differences between cursor movements at the segment level. Evaluation results on two mouse-behavior datasets show that MAUSPAD yields the best overall performance among tested schemes, and demonstrate the effectiveness of PADTW over DTW, and segmentation over non-segmentation. The processing techniques developed herein can be extended to applications that rely on sequence comparison, and where relevant sequence information spans multiple semantic domains.

The restriction of access to software systems is more important than ever. For example, critical data is increasingly being stored on web services that are accessible from anywhere in the world. Yet most primary authentication methods are still largely based on passwords, which are vulnerable to various attacks such as phishing scams and keyloggers. Advanced methods of behavior-based authentication exist, but most are designed for a specific area or system and are not generally applicable. In this paper, we propose a generic continuous authentication scheme for software systems, which supplements existing authentication schemes and works as an auxiliary layer to provide additional protection against impostors. The kernel of our scheme is a novel monitoring engine that detects impostors in real-time based on behavior and context information. We evaluate our scheme on a dataset consisting of real users' historical records provided by our industrial partner, and the results demonstrate that our approach achieves a high classification accuracy with only a short delay in detection, allowing for real-time, continuous authentication.

In this paper, a video abstract system based on spatial-temporal neighborhood trajectory analysis algorithm which is mainly used to process surveillance videos is proposed. The algorithm uses the spatial adjacency of foreground targets and tracks the spatial-temporal neighboring moving targets to get their whole trajectories in order to meet the requirement of processing speed and accuracy. The indicators consist of trajectory detection rate, trajectory tracking average continuity and video abstract processing speed are used to evaluate the effectiveness of the system. We compare the algorithm with the other three algorithms, and the results show that spatial-temporal neighborhood trajectory analysis algorithm has sufficient trajectory detection rate and processing speed for surveillance video abstraction.