edge computing

EVA: Efficient Token Compression Co-Optimization With Heterogeneous CIM Architecture for Video LLM Acceleration

Video large language models (VLLMs) enable powerful multimodal reasoning but face severe efficiency challenges on edge devices due to the massive computational and memory demands caused by lengthy video-token sequences. Existing methods struggle to efficiently compress spatiotemporally redundant tokens while minimizing DRAM access overhead during inference. In this work, we propose EVA, a co-designed algorithm–hardware framework for accelerating VLLMs. At the algorithm level, we introduce an efficient training-free token compression (ETC) method that combines greedy temporal segmentation (GTS) to adaptively partition frames by content similarity, with diversity spatiotemporal compression (DSC) to retain semantically rich tokens from both static and dynamic regions. The method is plug-and-play, requires no retraining, and employs sign similarity to enable hardware-friendly computing at scale. At the hardware level, we design a heterogeneous accelerator integrating a lightweight token compression engine (TCE), a computing-in-memory (CIM) engine for in situ execution of linear layers, and a reconfigurable digital attention engine with hardware-specialized attention computation and an interleaved pipeline dataflow. Across multiple models and video benchmarks, EVA preserves accuracy under aggressive token reduction and delivers substantial efficiency gains. Notably, on LLaVA-OneVision-7B, EVA compresses 90% of video tokens while maintaining 97.8% of the original accuracy. EVA achieves up to <inline-formula> <tex-math notation="LaTeX">$10.3\times $ </tex-math></inline-formula> speedup and <inline-formula> <tex-math notation="LaTeX">$47.4\times $ </tex-math></inline-formula> energy reduction compared to GPU, and achieves up to <inline-formula> <tex-math notation="LaTeX">$3.8\times $ </tex-math></inline-formula> speedup over prior specialized accelerators, demonstrating a practical path toward scalable VLLM inference on resource-constrained devices.

Edge-Girth as a Structural Edge Feature for Graph Neural Networks

Graph neural networks (GNN) based on message passing are provably no more powerful than the one-dimensional Weisfeiler--Leman colour-refinement test (1-WL): two graphs it cannot tell apart receive identical representations, however deep or wide the network. A common remedy augments node or edge features with precomputed structural descriptors, most often counts of a fixed small subgraph such as triangles or longer cycles, but such counts require committing in advance to the size of the substructure counted, a choice usually made blind to the data. We study a descriptor that avoids this choice. The edge-girth of an edge is the length of a shortest cycle through it, and its multiplicity is the number of such shortest cycles; together they form a per-edge invariant that reports cycles of arbitrary length, computable exactly by a single breadth-first search per edge. Injected into a gated message-passing architecture, EGAGNN, it reaches a test MAE a factor three below the closest gated comparator on the ZINC-12k regression benchmark at 104k parameters; against bounded cycle-counting descriptors under the same architecture, it matches only a dictionary counting cycles up to length eight, using twice as many channels, while a dictionary capped at length four performs no better than no structural information at all. On graph discrimination we prove a matching limitation: on graphs where every edge sees the same number of shortest cycles of the same length, the descriptor becomes constant and any model built on it collapses back to the 1-WL bound. This holds without exception across all 400 pairs of the BREC benchmark: not one of the 90 such pairs is distinguished.

Evaluating Multimodal LLMs as Generalist Vision-Language-Action Agents for Drone Control: Commanding, Approaching, Tracking and Searching

Multimodal Large Language Models (MLLMs) are strong perceivers of images and video. We ask how far that reach extends into acting: dropping an MLLM directly into a drone's control loop, with its entire action space declared solely in the prompt. Recent systems approach this setting but increasingly narrow the model's decision-making. We widen it back. We introduce DroneCATS-Agent, an architecture where the MLLM is a swappable component, and DroneCATS, a benchmark treating the model as the independent variable. Beyond merely flying toward a pixel, our agent entrusts the model to yaw and search, deliberate when unsure, and self-declare arrival---all without fine-tuning or function-calling schemas. Evaluating frontier and open models across four core capabilities---approaching a visible target, tracking a moving one, searching outside the initial view, and commanding a multi-drone fleet---reveals that even the simplest embodied settings are far from solved. Crucially, to identify what breaks first at the edge, our roster scales down to 2B parameters. The findings expose a stark paradox: it is not the flying that fails. Small open models often navigate into the success radius more reliably than frontier models, yet lose the episode by declaring arrival prematurely or not at all. Multi-drone commanding amplifies this divide, with small models failing by blindly copying a single coordinate across distinct views. Viewed as vision-language-action agents, the models'spatial perception holds up, but their action protocol does not. What separates a deployable edge model from a frontier model is not navigation, but the discipline to sustain a declared protocol and emit the correct terminating action. The open problem is closing this gap at onboard compute costs---yielding a fast model that plans persistently and knows exactly when it is done---and DroneCATS is built to measure that distance.

Architecturale tegenmaatregelen tegen hardwareaanvallen op RISC-V-kernen

The goal of this research is to investigate and develop embedded architectures for edge computing applications. The project targets HW/SW hybrid platforms, composed of FPGA logic coupled with embedded processors, and focuses on two aspects which are critical to IoT sensor-based settings. On the one hand, it involves the design of tailored hardware accelerators exploiting the parallel processing and hardware reconfiguration properties of FPGAs, in order to improve the performance of computationally-intensive tasks. Such tasks include signal processing or cryptographic algorithms, which are often found in edge nodes that collect and process sensor-based data. On the other hand, the project also investigates the design of processor architectures hardened against information leakage. It investigates methods to safeguard software executing in the edge platforms from adversaries aiming to break their security guarantees. The research activities will be performed around extensible processor architectures such as RISC-V. A central question is to determine the impact of the extensions in terms of cost, performance and energy consumption w.r.t. baseline designs.

A Simple Algorithm for the Directed Multiple Source Replacement Paths Problem

In the replacement paths (RP) problem, we are given a graph $G = (V, E)$ with $n = |V|$ and $m = |E|$, together with two vertices $s, t \in V$, and are asked to compute the shortest-path distance from $s$ to $t$ in $G \setminus e$ for every failed edge $e \in E$. The multiple source replacement paths (MSRP) problem is its natural generalization: given a set $S \subseteq V$ of $\sigma$ sources, compute the replacement path distances for all pairs in $S \times V$. In this paper, we present a randomized combinatorial algorithm that solves MSRP on unweighted directed graphs in $\tilde{O}(m\sqrt{\sigma n} + \sigma n^2)$ time, with all the output distances correct with high probability. This improves the best known bound $\tilde{O}(m\min\{\sigma\sqrt{n}, n\} + \sigma n^2)$ for directed graphs, which is obtained either by running the single source RP algorithm of Chechik and Magen [ICALP'20] from each source separately or by constructing and querying the all-pairs distance sensitivity oracle of Bernstein and Karger [STOC'09]. Our running time is essentially tight among combinatorial algorithms because Gupta, Jain, and Modi [PODC'20] proved a lower bound of $m{(\sigma n)}^{1/2-o(1)}$ for such algorithms, which holds even on undirected graphs, and the additive term $\sigma n^2$ is proportional to the time needed to write down the $\Theta(\sigma n^2)$ output distances. The algorithm is also remarkably simple.

CDF-DETR: Cross-Stage Attention and Dual-Scale Feature Calibration for Small-Object Detection in UAV Remote Sensing Imagery

Small-object detection in unmanned aerial vehicle (UAV) remote sensing imagery is challenged by dense target distributions, substantial scale variation, complex ground backgrounds, and limited edge-computing resources. To address these challenges, we propose CDF-DETR, an end-to-end detector derived from the Real-Time Detection Transformer (RT-DETR). First, a Cross-Stage Partial Single-Head Attention Transformer (CSP-SHAT) backbone combines efficient local feature extraction with partial-channel global interaction to improve multi-scale representation while reducing the parameter count of the backbone. Second, a dual-scale feature calibration (DSFC) module sequentially performs contextual aggregation and deformable spatial alignment, thereby improving the consistency of shallow localization features and deep semantic features. Third, Focaler-MPDIoU integrates coordinate-sensitive regression with IoU-quality-based sample reweighting for dense small-object localization. Experiments on the VisDrone-2019 test set and the UAVDT and HIT-UAV validation sets demonstrate mAP50 improvements of 3.1, 1.4, and 3.0 percentage points, respectively, over the RT-DETR-R18 baseline. On the VisDrone-2019 validation set, CDF-DETR improves mAP50–95 from 26.20% to 28.52%, corresponding to a gain of 2.32 percentage points, while reducing the parameter count by 25.7%. A compressed INT8 variant achieves 20.84 FPS for an offline image-level pipeline on an NVIDIA Jetson Orin Nano using ONNX and TensorRT. These results demonstrate improved detection accuracy with a reduced parameter footprint for UAV remote sensing image analysis.

Research on Edge-Cloud Collaborative Resource Scheduling and Security Management Based on Intelligent Optimization and Privacy Protection

With the rapid development of edge computing, cloud computing, big data, artificial intelligence and the Internet of Things, traditional centralized cloud computing faces increasing challenges in latency, bandwidth pressure, resource utilization and data privacy protection. Edge-cloud collaboration provides a new computing paradigm by extending computing, storage and network resources from centralized cloud data centers to edge nodes closer to users and data sources. This architecture can improve service response efficiency and reduce data transmission pressure, but it also introduces heterogeneous resources, dynamic task requests, unclear security boundaries and privacy leakage risks. This paper reviews existing studies on edge-cloud collaboration, resource scheduling, intelligent optimization and privacy protection, and then conducts an analytical discussion of the research gaps rather than an experimental evaluation. The review shows that existing studies have achieved valuable progress in task offloading, resource allocation, deep reinforcement learning, access control, encryption and federated learning. However, research on integrated frameworks that combine intelligent resource scheduling with privacy-aware security management remains limited. To address this limitation, this paper proposes a formalized privacy-aware scheduling perspective that incorporates latency, energy consumption, cost, node trustworthiness, data sensitivity, privacy leakage risk, reliability and auditability into a unified decision model. The analysis indicates that future edge-cloud systems should evolve from efficiency-centric scheduling toward secure, trustworthy and sustainable collaborative governance.

Leveraging Cyber-Physical Security Solutions Blended With Machine Learning for Advanced IoT Botnet Detection

The proliferation of Internet of Things (IoT) ecosystems has significantly increased the attack surface of cyber-physical systems, leading to the emergence of large-scale botnets that exploit device vulnerabilities for distributed and persistent attacks. This survey comprehensively reviews state-of-the-art techniques that integrate cyber-physical security solutions with machine learning (ML) for advanced detection and mitigation of IoT botnets. It categorizes existing methods into networkcentric, host-based, and hybrid cyber-physical detection frameworks, emphasizing their detection granularity, scalability, and computational feasibility in resource-constrained environments. The paper systematically analyzes supervised, unsupervised, and deep learning approaches, ranging from Random Forests and Support Vector Machines to Autoencoders, LSTMs, and Graph Neural Networks, highlighting their adaptability to evolving botnet behaviours and zero-day threats. Furthermore, the survey explores the integration of federated learning, edge computing, and software-defined networking (SDN) to enable distributed, privacy-preserving, and realtime detection architectures. Key challenges, including data imbalance, adversarial resilience, explainability, and cross-domain generalization, are critically discussed. Finally, this work outlines a taxonomy of cyber-physical and ML-based IoT botnet detection models and identifies future research directions toward autonomous, adaptive, and explainable cyber-physical defense systems.

LumiCharge: Spherical Harmonic Convolutional Networks for Atomic Charge Prediction in Drug Discovery.

Atomic charge is crucial in drug design for analyzing reactive sites and interactions between ligands and targets. While quantum mechanical methods offer high accuracy, they are generally computationally costly. Conversely, empirical approaches, while computationally efficient, frequently suffer from lack of precision and generalizability. Recent a number of machine learning-based models have been developed for atomic charge predictions, but they struggle with accurately representing molecular structures and capturing the chemical environments affecting atomic charges, thus limiting their generalization and accuracy. To overcome these limitations, we propose LumiCharge, a novel atomic charge prediction framework that incorporates high-order spherical harmonics convolutions and explicitly models multibody interactions. In constructing this model, we employ a strategy that integrates both high- and low-order information, enhancing its geometric spatial perception capability, which is currently underexplored in the field. Benchmark evaluations demonstrate that LumiCharge outperforms state-of-the-art (SOTA) models by 30%-60% across diverse data sets. Additionally, in cross-scale experiments, LumiCharge demonstrates exceptional extrapolation capability and robustness across molecules of varying sizes, effectively overcoming the limitations imposed by molecular sizes. On an external halogen-containing test set, LumiCharge achieves an RMSE of 0.055e, meeting practical application requirements. Finally, a case study of virtual screening for the androgen receptor (AR) target further validates its outstanding accuracy compared to the OPLS3e force field and other deep learning (DL)-based baseline models, highlighting its exceptional generalization capacity and practical utility in real-world scenarios.

PepBAN: A Deep Learning Framework with Bilinear Attention and Adversarial Learning for Peptide-Protein Interaction Prediction

Accurate prediction of the peptide-protein interaction (PepPI) is crucial for developing peptide-based therapeutics and vaccines. However, this computational task has traditionally faced significant challenges, such as the scarcity of structure data along with the corresponding label of the binding affinity for bound complexes. To address these challenges, we introduce PepBAN, a deep learning framework for modeling PepPI predictions. PepBAN incorporates two technical advancements: (1) adopting the protein language model ESM-2 to characterize proteins and ESM-2 or a graph-based foundation model for peptides without structure data and (2) leveraging the conditional domain adversarial learning to enhance generalization across a broad range of protein targets, especially when there are limited binding data. At the core of PepBAN is a bilinear attention network (BAN) that effectively learns the pattern of pairwise local interactions, enables the identification of key residues participating in the peptide-protein interactions, and offers an intuitive approach to interpret the underlying mechanisms of PepPIs via analyzing attention weights. Our numerical experiments demonstrated that PepBAN outperformed the previous state-of-the-art models across several well-established benchmark studies. Furthermore, we evaluated PepBAN's applicability in predicting cyclic peptide-protein interactions, a task that poses significant challenges due to the presence of noncanonical amino acids. These nonstandard residues require specialized handling, which most existing sequence-based PepPI prediction models did not adequately address, and we adopt an atom-resolved molecular graph approach to process cyclic peptides. Despite this complexity, PepBAN demonstrated a clear advantage by achieving a superior prediction performance and offering a distinct edge in tackling the emerging chemical space of cyclic peptides, which has great potential for novel therapeutic development. In summary, PepBAN serves as a valuable tool for advancing peptide-based drug and therapeutic development.

SynGFN: learning across chemical space with generative flow-based molecular discovery

Allosteric Cooperativity Mechanism Investigation of Orthosteric and Allosteric Ligands in Modulating AR Activity: A Molecular Dynamics Study

The androgen receptor (AR) represents a pivotal therapeutic target for prostate cancer. However, existing orthosteric ligand-binding pocket (LBP) antagonists [e.g., enzalutamide (ENZ)] encounter significant obstacles due to resistance-conferring mutations in the LBP. Allosteric antagonists targeting the BF3 site exhibit great potential in overcoming such resistance but have low inhibitory efficacy. In our study, we employed an integrated computational modeling strategy, including Gaussian-accelerated molecular dynamics (GaMD), MM/GBSA free-energy calculations, and elastic network model (ENM)-based signaling communication pathway analyses. This approach is used to probe the cooperativity of allosteric BF3 antagonists [e.g., VPC-13808 (VPC)] with diverse orthosteric LBP ligands [e.g., ENZ and testosterone (TES)] in suppressing AR activity. Herein, four types of AR systems were examined: AR bound to LBP agonist (AR·TES), LBP antagonists (e.g., AR·ENZ), and combinations of LBP agonist/antagonist with BF3 antagonist (e.g., AR·TES·VPC and AR·ENZ·VPC). Results indicate that BF3 antagonists can synergize with the LBP antagonist to amplify conformational flexibility in H12 and induce anticorrelated dynamics of H12 with H3 and H4. This induces the downward movement of H12 and its displacement away from H3/H4, triggering the wide opening of the AF2 binding cleft and substantially reducing the coactivator recruitment. Furthermore, the BF3 antagonist can interact with specific residues (e.g., F673, F826, L830, and Y834) and cooperate with the LBP agonist or antagonist to allosterically perturb the AF2 conformation. Multiple short- and/or long-range BF3→AF2 and LBP→AF2 signaling transition pathways are involved, such as F673→Y834→L722→L812→L744→V746→L873→ENZ→L880/V889/V891. These mechanistic insights establish the foundation for developing novel AR BF3 antagonist and LBP-BF3 combination therapies, suggesting a promising avenue for enhancing the efficacy and overcoming the resistance in castration-resistant prostate cancer treatment.