Although large language models (LLMs) have demonstrated remarkable code-generation ability, they still struggle with complex tasks. In real-world software development, humans usually tackle tasks through collaborative teamwork, a strategy that significantly controls development complexity and enhances quality. Inspired by this, we present self-collaboration framework for code generation employing LLMs, exemplified ChatGPT. Specifically, role instructions, (1) Multiple LLM agents act as...

Signal recognition is one of the significant and challenging tasks in signal processing communications field. It often a common situation that there's no training data accessible for some classes to perform task. Hence, as widely-used image field, zero-shot learning (ZSL) also very important recognition. Unfortunately, ZSL regarding this field has hardly been studied due inexplicable semantics. This paper proposes framework, reconstruction convolutional neural networks (SR2CNN), address...

Although large language models (LLMs) have demonstrated impressive ability in code generation, they are still struggling to address the complicated intent provided by humans. It is widely acknowledged that humans typically employ planning decompose complex problems and schedule solution steps prior implementation. To this end, we introduce into generation help model understand reduce difficulty of problem-solving. This paper proposes a self-planning approach with models, which consists two...

Due to the emergence of deep learning, signal recognition has made great strides in performance improvement. The success most learning methods relies on accessibility abundant labeled training data. However, annotation signals is quite expensive, making it challenging train models substantially. This calls for development semi-supervised (SSL) method fully utilize unlabeled data assist models. To achieve this goal, three types loss functions, tailored task SLL-based recognition, are...

A proper code evaluation metric (CEM) profoundly impacts the evolution of generation, which is an important research field in NLP and software engineering. Prevailing match-based CEMs (e.g., BLEU, Accuracy, CodeBLEU) suffer from two significant drawbacks. 1. They primarily measure surface differences between codes without considering their functional equivalence. However, equivalence pivotal evaluating effectiveness as different can perform identical operations. 2. are predominantly designed...

Unmanned aerial vehicle (UAV) communication has drawn significant interests from the industry and academic due to its low cost, high maneuverability, on-demand deployment. This paper considers priority-based resource coordination in a multi-UAV system, where multiple UAVs are operated by ground base station perform certain task with pre-defined trajectory. To ensure reliable control of UAVs, we formulate problem as mixed-integer programming, mainly aiming maximize minimum signal-to-noise...

Deep neural networks have been shown as a class of useful tools for addressing signal recognition issues in recent years, especially identifying the nonlinear feature structures signals. However, this power most deep learning techniques heavily relies on an abundant amount training data, so performance classic nets decreases sharply when number data samples is small or unseen are presented testing phase. This calls advanced strategy, i.e., model-agnostic meta-learning (MAML), which able to...

Code generation aims to automatically generate the source code based on given natural language (NL) descriptions, which is of great significance for automated software development. Some models follow a model-based paradigm (LMBP) tokens sequentially. others focus deriving grammatical structure by generating program's abstract syntax tree (AST), i.e., using structure-based (GSBP). Existing studies are trying through one above two models. However, human developers often consider both...

Due to the emergence of deep learning, signal recognition has made great strides in performance improvement. The success most learning methods relies on accessibility abundant labelled training data. However, annotation signals is quite expensive, making it challenging train models substantially. This calls for development semi-supervised (SSL) method fully utilize unlabelled data assist models. To achieve this goal, three types loss function tailored task are carefully designed paper....

Deep neural networks have been shown as a class of useful tools for addressing signal recognition issues in recent years, especially identifying the nonlinear feature structures signals. However, this power most deep learning techniques heavily relies on an abundant amount training data, so performance classic nets decreases sharply when number data samples is small or unseen are presented testing phase. This calls advanced strategy, i.e., model-agnostic meta-learning (MAML), which can...

Current structured pruning methods often result in considerable accuracy drops due to abrupt network changes and loss of information from pruned structures. To address these issues, we introduce the Decay Pruning Method (DPM), a novel smooth approach with self-rectifying mechanism. DPM consists two key components: (i) Smooth Pruning: It converts conventional single-step into multi-step pruning, gradually reducing redundant structures zero over N steps ongoing optimization. (ii)...

Different from the flow semantics of natural languages, programming languages are inherently rigid in structure and grammar. Existing fine-tuning methodologies for code vulnerability detection generally treat as long text sequences, stripping away structural elements such newlines ('/n') whitespace. However, this approach inadvertently results loss crucial information, diminishing distinct characteristics impairing accuracy detection. To address these challenges, we propose a novel network...

Large language models (LLMs) have achieved impressive performance in code generation recently, offering programmers revolutionary assistance software development. However, due to the auto-regressive nature of LLMs, they are susceptible error accumulation during generation. Once an is produced, LLMs can merely continue generate subsequent conditioned on it, given their inability adjust previous outputs. Existing LLM-based approaches typically consider post-revising after generation, leading...

The rotation averaging problem is a fundamental task in computer vision applications. It generally very difficult to solve due the nonconvex constraints. While sufficient optimality condition available literature, there lack of fast convergent algorithm achieve stationary points. In this paper, by exploring structure, we first propose block coordinate descent (BCD)-based with guaranteed convergence Afterwards, further an alternative applying successive upper-bound minimization (SUM) method....