Multi-task large language models (MTLLMs) are important for many applications at the wireless edge, where users demand specialized models to handle multiple tasks efficiently. However, training MTLLMs is complex and exhaustive, particularly when tasks are subject to change. Recently, the concept of model fusion via task vectors has emerged as an efficient approach for combining fine-tuning parameters to produce an MTLLM. In this paper, the problem of enabling edge users to collaboratively craft such MTLMs via tasks vectors is studied, under the assumption of worst-case adversarial attacks. To this end, first the influence of adversarial noise to multi-task model fusion is investigated and a relationship between the so-called weight disentanglement error and the mean squared error (MSE) is derived. Using hypothesis testing, it is directly shown that the MSE increases interference between task vectors, thereby rendering model fusion ineffective. Then, a novel resilient MTLLM fusion (R-MTLLMF) is proposed, which leverages insights about the LLM architecture and fine-tuning process to safeguard task vector aggregation under adversarial noise by realigning the MTLLM. The proposed R-MTLLMF is then compared for both worst-case and ideal transmission scenarios to study the impact of the wireless channel. Extensive model fusion experiments with vision LLMs demonstrate R-MTLLMF's effectiveness, achieving close-to-baseline performance across eight different tasks in ideal noise scenarios and significantly outperforming unprotected model fusion in worst-case scenarios. The results further advocate for additional physical layer protection for a holistic approach to resilience, from both a wireless and LLM perspective.

Despite recent advancements in millimeter-wave (mm Wave) communication systems, the practical validation of research ideas in this field is still challenging. This is mostly due to the simplifying assumptions in their system modeling and overlooking hardware limitations in the simulation environment. In this paper, we propose a high-speed and programmable mm Wave testbed with multi-user (MU) multiple-input multiple-output (MIMO) beamforming capability. The baseband signal processing is preformed fully on a general-purpose processor. The radio frequency (RF) frontend holds a hybrid MIMO architecture, where each RF chain is connected to one phased-array antenna. The testbed allows to plug and play different physical layer signal processing algorithms or radio resource allocation techniques, including machine learning models, to assess their performance under real-life conditions.

Edge computing is a sustainable network paradigm that supports resource constrained Industrial Internet of Things (IIoT) devices in performing data-driven tasks. The distributed architecture of computing resources, combined with the heterogeneous and sporadic characteristics of IIoT tasks, presents a substantial challenge to achieving energy-efficient and low-latency task execution. In this paper, we demonstrate that modeling the Age of Information (AoI) as an exponential function can significantly enhance the performance of task scheduling and execution in edge computing systems. Furthermore, we introduce a low over-head Multi-Agent Deep Reinforcement Learning (DRL) algorithm with No Observation (MA-NO) for distributed edge computing resource allocation, where agents (IIoT devices) operate independently without communicating with each other. However, to prevent the divergence of the agents and ensure a Pareto-optimal decision-making, we adopt a common reward that is shared among all agents. That is, all agents receive the same reward, based on their collective performance. Simulation results show that our proposed AoI model reduces the average execution latency of the tasks by 35.2% compared to a linear AoI ones. Furthermore, our proposed MA-NO algorithm reduces the total energy consumption of the IIoT devices by 59.9% and 63.4% compared to the single-agent DRL algorithm and multi-agent DRL algorithm with partial observations, respectively.

We propose an extremely energy efficient and reconfigurable accelerator for performing Deep Neural Network (DNN) inferences on a Field-Programmable Gate Array (FPGA). Our design allows on the fly reconfigurability of the model to adapt it to any arbitrary DNN inference task and perform it with an extremely low latency, on the scale of tens of micro seconds. Our design can be adopted by resource constrained Industrial Internet of Things (IIoT) devices (e.g., mobile robots) to fulfill different DNN inference tasks. By parallelization of the matrix multiplications in each layer of the DNN and adopting an adjustable sliding window at the synthesize stage, our accelerator strikes a balance between the computing latency and the scarce resource utilization on the FPGA. It also uses a low-overhead control scheme for the accelerator’s internal interactions to update the DNN model on-the-fly. The results show that for a DNN model with 64 input features, 9 hidden layers (each with 64 neurons), and 64 output variables, our proposed inference accelerator speeds up the inference by 28.8 times and decreases the energy consumption by 31 % compared to an AMD Deep Learning Processing Unit.

Integrating distributed machine learning tools with sensing, communication, and decision-making operations in networked and cooperative intelligent machines has opened up novel avenues for research. The cross-fertilization of these components is essential for enabling collaborative task management that requires safety, reliability, scalability, and low latency. Data privacy preservation and high spectral efficiency are required for various Industrial Internet of Things (IIoT) applications. Most previous works have focused on joint optimization of task-offloading and resource allocation while neglecting the data privacy and data overhead of the decision-making process in task-offloading. Considering the robos as agents, this paper proposes Prioritized Spectral efficient Federated Reinforcement Learning (PSFRL)-based partial task-offloading in a heterogeneous industrial robotic Mobile Edge Computing (MEC) system. Due to its definition, PSFRL keeps data private while achieving high spectrum efficiency. Additionally, we define a Value of Information (VoI) metric so that the PSFRL agents allocate their resources more wisely for more valuable data. As robots have limited battery and computing resources for task processing, multiple edge computing-enabled access points assist the robots in accomplishing tasks. Simulation results show that our proposed PSFRL outperforms other learning-based task-offloading methods concerning energy consumption, spectral efficiency, and VoI. Moreover, we demonstrate the superiority of the proposed cooperative approach over its non-cooperative counterpart.

In delay-sensitive industrial Internet of Things (IIoT) applications, the age of information (AoI) is employed to characterize the freshness of information. Meanwhile, the emerging network function virtualization provides flexibility and agility for service providers to deliver a given network service using a sequence of virtual network functions (VNFs). However, suitable VNF placement and scheduling in these schemes is NP-hard and finding a globally optimal solution by traditional approaches is complex. Recently, deep reinforcement learning (DRL) has appeared as a viable way to solve such problems. In this paper, we first utilize single agent low-complex compound action actor-critic RL to cover both discrete and continuous actions and jointly minimize VNF cost and AoI in terms of network resources under end-to-end Quality of Service constraints. To surmount the single-agent capacity limitation for learning, we then extend our solution to a multi-agent DRL scheme in which agents collaborate with each other. Simulation results demonstrate that single-agent schemes significantly outperform the greedy algorithm in terms of average network cost and AoI. Moreover, multi-agent solution decreases the average cost by dividing the tasks between the agents. However, it needs more iterations to be learned due to the requirement on the agents' collaboration.