Within the realm of software engineering, specialized tasks on code, such as program repair, present unique challenges, necessitating fine-tuning Large language models (LLMs) to unlock state-of-the-art performance. Fine-tuning approaches proposed in the literature for LLMs on program repair tasks generally overlook the need to reason about the logic behind code changes, beyond syntactic patterns in the data. High-performing fine-tuning experiments also usually come at very high computational costs. With MORepair , we propose a novel perspective on the learning focus of LLM fine-tuning for program repair: we not only adapt the LLM parameters to the syntactic nuances of the task of code transformation (objective ➊), but we also specifically fine-tune the LLM with respect to the logical reason behind the code change in the training data (objective ➋). Such a multi-objective fine-tuning will instruct LLMs to generate high-quality patches. We apply MORepair to fine-tune four open-source LLMs with different sizes and architectures. Experimental results on function-level and repository-level repair benchmarks show that the implemented fine-tuning effectively boosts LLM repair performance by 11.4% to 56.0%. We further show that our fine-tuning strategy yields superior performance compared to the state-of-the-art approaches, including standard fine-tuning, Fine-tune-CoT, and RepairLLaMA.

Mobile edge computing (MEC) has become an effective paradigm to support computation-intensive applications by providing services in close proximity to user devices (UDs). In MEC networks, computation offloading technology is devoted to balancing system load and prolonging UDs' battery life. However, most existing studies on computation offloading take the impractical assumption of the MEC scenario with homogeneous users, ignoring security requirement from certain users. Moreover, with users mobility and task arrivals correlation, most existing computing offloading approaches suffer from inefficient or suboptimal decision making in practical MEC environments. To tackle these issues, by integrating task arrivals correlation within a time slot and environment dynamics between time slots, we propose an adaptive computation offloading scheme based on a collaborative architecture with heterogeneous MEC nodes. First, considering additional security requirement from very important people (VIP) users, we present a novel collaborative architecture by separating edge/cloud servers into public and private nodes. Then, with the architecture, we develop a dynamic computation offloading (DCO) algorithm to realize adaptive computation offloading scheme in MEC environment with mobile users. Particularly, the algorithm involves three stages. 1) By extending Poisson process into Markovian arrival process (MAP), we construct an MAP-based system model to capture the behavior of time-dependent task arrivals and then analyze the system model to derive the system delay in steady state. 2) For the purpose of minimizing the system delay in each time slot, we formulate a computation offloading problem in MEC environment with mobile users. 3) Under a deep reinforcement learning (DRL) framework, by taking the system delay as environmental feedback, we solve the formulated problem and provide offloading decisions in each time slot. We evaluate the performance of DCO algorithm by comparing it with other benchmark algorithms in various application scenarios. Results demonstrate that the proposed DCO algorithm outperforms the compared algorithms in response performance.

With the proven effectiveness of Large Language Models (LLMs) in code-related tasks, researchers have explored their potential for program repair. However, existing repair benchmarks might have influenced LLM training data, potentially causing data leakage. To evaluate LLMs’ realistic repair capabilities, (i) we introduce an extensive, non-crawled benchmark TutorCode, comprising 1,239 C++ defect codes and associated information such as tutor guidance, solution description, failing test cases, and the corrected code. Our work assesses LLM’s repair performance on TutorCode, measuring repair correctness (TOP-5 and AVG-5) and patch precision (RPSR). (ii) We then provide a comprehensive investigation into which types of extra information can help LLMs improve their repair performance. Among these types, tutor guidance was the most effective information. To fully harness LLMs’ conversational capabilities and the benefits of augmented information, (iii) we introduce a novel conversational semi-automatic repair framework CREF assisting human programming tutors. It demonstrates a remarkable AVG-5 improvement of 17.2%-24.6% compared to the baseline, achieving an impressive AVG-5 of 76.6% when utilizing GPT-4. These results highlight the potential for enhancing LLMs’ repair capabilities through tutor interactions and historical conversations. The successful application of CREF in a real-world educational setting demonstrates its effectiveness in reducing tutors’ workload and improving students’ learning experience, showing promise for code review and other software engineering tasks.

How to collaboratively offload tasks between user devices, edge networks (ENs), and cloud data centers is an interesting and challenging research topic. In this paper, we investigate the offloading decision, analytical modeling, and system parameter optimization problem in a collaborative cloud-edge-device environment, aiming to trade off different performance measures. According to the differentiated delay requirements of tasks, we classify the tasks into delay-sensitive and delay-tolerant tasks. To meet the delay requirements of delay-sensitive tasks and process as many delay-tolerant tasks as possible, we propose a cloud-edge-device collaborative task offloading scheme, in which delay-sensitive and delay-tolerant tasks follow the access threshold policy and the loss policy, respectively. We establish a four-dimensional continuous-time Markov chain as the system model. By using the Gauss-Seidel method, we derive the stationary probability distribution of the system model. Accordingly, we present the blocking rate of delay-sensitive tasks and the average delay of these two types of tasks. Numerical experiments are conducted and analyzed to evaluate the system performance, and numerical simulations are presented to evaluate and validate the effectiveness of the proposed task offloading scheme. Finally, we optimize the access threshold in the EN buffer to obtain the minimum system cost with different proportions of delay-sensitive tasks.

In vehicular edge computing (VEC) networks, vehicle terminal (VT) typically offloads tasks to road side units (RSUs) equipped with edge servers to obtain service with low latency. However, the lack of global information and time-varying nature of VEC networks present challenges to make effective resource allocation decisions under long-term constraints. Motivated by this, we aim to investigate a dynamic resource allocation scheme with diverse tasks. We formulate an optimization problem to minimize average task delay under long-term constraints of energy consumption and system cost for cloud-edge collaboration offloading. For the coupling of resource allocation decisions between different time slots, we propose a Lyapunov online resource allocation (LORA) algorithm. LORA transforms the formulated problem into a problem of minimizing the upper bound of the drift-plus-penalty function. We then decompose the latter into multiple subproblems and provide the corresponding algorithms for solving them separately. Experimental results show that our proposed LORA reduces energy consumption and system cost by 10.1% and 4.2%, respectively, compared to DO, and reduces average task delay by 15.1% compared to ECSCO.

Utilizing traditional object detectors to build a few-shot object detection (FSOD) model ignores the differences between classification and regression tasks and causes task conflict and class confusion, resulting in a decline in classification performance. In contrast, this paper focuses on the above shortcomings and utilizes the strategies of Back Propagation and Dynamic Learning to construct a model for addressing FSOD, named BPDL. Our BPDL has a two-fold main idea: a) it uses the optimized localization boxes to alleviate the task conflict and refine classification features by a correction loss, and b) it develops a dynamic learning strategy to filter the confusing features and mine more realistic prototype representations of the categories to calibrate classification. Extensive experiments on multiple benchmarks show that our BPDL model outperforms existing methods and advances the FSOD task’s state-of-the-art.

Summary With the development of computing technology and the popularization of application in 5G network, mobile‐edge computing (MEC), which can effectively reduce time delay and save energy consumption, has attracted extensive attention from the academic community. How to make a suitable allocation decision for tasks becomes one of the critical issues in MEC systems. In order to yield the most benefit to a newly arriving task, by placing a decision making module (DMM) in a mobile device, a MEC system architecture is presented. Based on the net benefit of a newly arriving task, the DMM makes a decision of dropping the task, allocating the task to the local execution system, or offloading the task to the MEC server via the transmission system. From the observable perspective of task individuals, a pure threshold strategy is proposed to show that a Nash equilibrium is always allowed. By constructing a two‐dimensional continuous‐time Markov chain, a social optimal threshold strategy is proposed. Numerical results show that threshold under the pure threshold strategy is always greater than that under the social optimal threshold strategy. For this, a charging policy is presented to coincide the pure threshold strategy with the social optimal threshold strategy.

Abstract The energy consumption of virtualized data centers has grown very fast in last several years. Because a large number of hosts are running in an idle state, virtualized data centers waste a large amount of electric energy. To save more energy for virtualized data centers, an energy‐saving mechanism is proposed based on switching operating mode of physical servers and reserving virtual machines (VMs). The main idea is that when the amount of idle VMs reaches twice of a specified threshold, half of these idle VMs are reserved to process the new task that is about to arrive, and the physical server which hosts the other half of the idle VMs is switched to the sleep mode. From the perspective of task arrival rate and sleep parameters, we use two‐dimensional Markov processes to analyze the proposed energy‐saving mechanisms. By using matrix geometry solutions, we theoretically estimate energy consumption and response performance. According to the numerical experiments, the proposed energy‐saving mechanism obviously cut down electric energy consumption and ensures response performance. Finally, the specified threshold for the amount of VMs is optimized by building a cost function.

Wireless regional area network (WRAN) adopts centralized network architecture and is currently one of the most typical cognitive radio networks. In order to reduce the energy consumption of the communication networks with the constraint of spectrum resource utilization, a working sleep mechanism is introduced into the base station (BS), and a novel energy saving strategy with dual rate transmission is proposed. Combining the multiple-vacation queue and priority queue, using the quasi-birth-death process and the matrix-geometric solution method, we assess the average latency and the forced termination probability of secondary user packets, as well as the energy saving ratio and the channel utilization of system. Based on the revenue-expenditure structure, a profit function is built, and then the Nash equilibrium behavior and the socially optimal behavior are investigated. With the help of the particle swarm optimization, an intelligent optimization algorithm to search the socially optimal arrival rate of secondary user packets is presented. In order to unify the arrival rates of secondary user packets with Nash equilibrium and social optimization, a reasonable pricing policy is formulated. In addition, system experiments are carried out to verify the effectiveness of the energy saving strategy and the rationality of the pricing policy.

In order to improve the spectrum efficiency and achieve greener communication in wireless applications, in this paper we propose a novel energy saving strategy in cognitive networks based on software defined radio. By establishing a preemptive priority queue model with single vacation, we capture the stochastic behavior of the proposed strategy. Using the method of matrix geometric solution, we derive both the average latency of secondary user packets and the energy saving rate. Finally, we provide numerical results to demonstrate the influence of the sleep timer length on the system performance, and to investigate the trade-off between different performance measures.

In this paper, we present the performance analysis of a digitized WSN by taking into account the switching procedure of the Phase-Locked Loop (PLL) from the sleep mode to the active mode. We model the digitized WSN as a discrete-time multiple vacation tandem multi-hop queueing network with a setup to capture the working principle of the sleep/wakeup protocol in IEEE 802.15.4. Correspondingly, for the performance measures, we present the formulas for the average latency of data frames and the total energy consumption of the system. Moreover, we present numerical results comparing with simulation results to discuss the relationship between the system performance and the number of multi-hops for different constellation sizes.

In this paper, we analyze the performance of automatic repeat request (ARQ) schemes for the service-oriented Internet in wireless networks. Considering the self-similar nature of a massive-scale multimedia service shown in Internet traffic, in this paper we present a discrete-time system operating on the basis of time slotting with a batch arrival. Moreover, taking into account the delay in the setting up procedure of the channels, we introduce a setup strategy in this system model. In the performance analysis, firstly, by using an imbedded Markov chain, we derive the probability generation functions (P.G.Fs.) of the queueing length, the waiting time and the busy period of the system. Then, we use these analyses to give some important performance measurements for wireless networks using ARQ schemes in terms of the response time and the setup ratio for the self-similar traffic. In numerical results, we make some comparisons such as the response time and the setup ratio with the systems by using other previously known ARQ schemes having the memoryless traffic to show the influence of the self-similar degree on the system performance.