Code Generation Research Articles

Developing GA-FuL: A Generic Wide-Purpose Library for Computing with Geometric Algebra

The Geometric Algebra Fulcrum Library (GA-FuL) version 1.0 is introduced in this paper as a comprehensive computational library for geometric algebra (GA) and Clifford algebra (CA), in addition to other classical algebras. As a sophisticated software system, GA-FuL is useful for practical applications requiring numerical or symbolic prototyping, optimized code generation, and geometric visualization. A comprehensive overview of the GA-FuL design is provided, including its core design intentions, data-driven programming characteristics, and extensible layered design. The library is capable of representing and manipulating sparse multivectors of any dimension, scalar kind, or metric signature, including conformal and projective geometric algebras. Several practical and illustrative use cases of the library are provided to highlight its potential for mathematical, scientific, and engineering applications. The metaprogramming code optimization capabilities of GA-FuL are found to be unique among other software systems. This allows for the automated production of highly efficient code, based on powerful geometric modeling formulations provided by geometric algebra.

DSCEH: Dual-Stream Correlation-Enhanced Deep Hashing for Image Retrieval

Deep Hashing is widely used for large-scale image-retrieval tasks to speed up the retrieval process. Current deep hashing methods are mainly based on the Convolutional Neural Network (CNN) or Vision Transformer (VIT). They only use the local or global features for low-dimensional mapping and only use the similarity loss function to optimize the correlation between pairwise or triplet images. Therefore, the effectiveness of deep hashing methods is limited. In this paper, we propose a dual-stream correlation-enhanced deep hashing framework (DSCEH), which uses the local and global features of the image for low-dimensional mapping and optimizes the correlation of images from the model architecture. DSCEH consists of two main steps: model training and deep-hash-based retrieval. During the training phase, a dual-network structure comprising CNN and VIT is employed for feature extraction. Subsequently, feature fusion is achieved through a concatenation operation, followed by similarity evaluation based on the class token acquired from VIT to establish edge relationships. The Graph Convolutional Network is then utilized to enhance correlation optimization between images, resulting in the generation of high-quality hash codes. This stage facilitates the development of an optimized hash model for image retrieval. In the retrieval stage, all images within the database and the to-be-retrieved images are initially mapped to hash codes using the aforementioned hash model. The retrieval results are subsequently determined based on the Hamming distance between the hash codes. We conduct experiments on three datasets: CIFAR-10, MSCOCO, and NUSWIDE. Experimental results show the superior performance of DSCEH, which helps with fast and accurate image retrieval.

EA4RCA: Efficient AIE accelerator design framework for regular Communication-Avoiding Algorithm

With the introduction of the Adaptive Intelligence Engine (AIE), the Versal Adaptive Compute Acceleration Platform (Versal ACAP) has garnered great attention. However, the current focus of Vitis Libraries and limited research has mainly been on how to invoke AIE modules, without delving into a thorough discussion on effectively utilizing AIE in its typical use cases. As a result, the widespread adoption of Versal ACAP has been restricted. The Communication Avoidance (CA) algorithm is considered a typical application within the AIE architecture. Nevertheless, the effective utilization of AIE in CA applications remains an area that requires further exploration. We propose a top-down customized design framework, EA4RCA (Efficient AIE accelerator design framework for regular Communication-Avoiding Algorithm), specifically tailored for CA algorithms with regular communication patterns, and equipped with AIE Graph Code Generator software to accelerate the AIE design process. The primary objective of this framework is to maximize the performance of AIE while incorporating high-speed data streaming services. Experiments show that for the RCA algorithm Filter2D and Matrix Multiple (MM) with lower communication requirements and the RCA algorithm FFT with higher communication requirements, the accelerators implemented by the RA4RCA framework achieve the highest throughput improvements of 22.19x, 1.05x and 3.88x compared with the current highest performance acceleration scheme (SOTA), and the highest energy efficiency improvements of 6.11x, 1.30x and 7.00x.

Automatic Code Generation for Android Applications Based on Improved Pix2code

With the expansion of the Internet market, the traditional software development method has been difficult to meet the market demand due to the problems of long development cycle, tedious work and difficult system maintenance. Therefore, to improve software development efficiency, this study uses residual networks and bidirectional long short-term memory networks to improve the Pix2code model. The experiment results show that after improving the visual module of the Pix2code model using residual networks, the accuracy of the training set improves from 0.92 to 0.96, and the convergence time is shortened from 3 hours to 2 hours. After using a bidirectional long short-term memory network to improve the language module and decoding layer, the accuracy and convergence speed of the model have also been improved. The accuracy of the training set grew from 0.88 to 0.92, and the convergence time was shortened by 0.5 hours. However, models improved by bidirectional long short-term memory networks might exhibit over-fitting, and thus this study uses Dropout and Xavier normal distribution to improve the memory network. The results validate that the training set accuracy of the optimized bidirectional long short-term memory network remains around 0.92, but the accuracy of the test set has improved to a maximum of 85%. Dropout and Xavier normal distributions can effectively improve the over-fitting problem of bidirectional long short-term memory networks. Although they can also decrease the model’s stability, their gain is higher. The training and testing accuracy of the Pix2code improved by residual network and bidirectional long short-term memory network are 0.95 and 0.82, respectively, while the code generation accuracy of the original Pix2code is only 0.77. The above data indicates that the improved Pix2code model has improved the accuracy and stability of code automatic generation.

Programmable delay cell based area efficient time-mode smart temperature sensor exploiting channel length modulation coefficient λ

This work presents an all-digital Programmable Delay Cell (PDC)-based time-domain smart temperature sensor exploiting the channel length modulation coefficient λ. For the first time the work has been described to establish the choice of multimode conversion options in the time-domain architecture of temperature sensor. The sensor incorporates a PDC and a time-to-digital converter (TDC) for the generation of temperature code. The different channel lengths of the PDC and Gated Ring Oscillator (GRO) delay elements of the TDC ensure the precise operation of the sensor. The proposed design provides a novel approach for the generation of variable resolution options having different sampling rates. At 90-nm CMOS process and 1.1 V supply voltage, the sensor occupies 0.015 mm2 and operates over a span of −60 °C to 120 °C. Additionally, multimode conversion time allows the sensor to be tuned to a small value of energy/conversion at 27 °C. This confirms to 0.38 nJ and 3.13 nJ at modes M0 and M7, respectively. Besides the sensor obtains a resolution 0.747 °C and 0.093 °C at a conversion time of 0.669 μs and 5.5 μs, respectively.

Rethinking AI code generation: a one-shot correction approach based on user feedback

Code generation has become an integral feature of modern IDEs, gathering significant attention. Notable approaches like GitHub Copilot and TabNine have been proposed to tackle this task. However, these tools may shift code writing tasks towards code reviewing, which involves modification from users. Despite the advantages of user feedback, their responses remain transient and lack persistence across interaction sessions. This is attributed to the inherent characteristics of generative AI models, which require explicit re-training for new data integration. Additionally, the non-deterministic and unpredictable nature of AI-powered models limits thorough examination of their unforeseen behaviors. We propose a methodology named One-shot Correction to mitigate these issues in natural language to code translation models with no additional re-training. We utilize decomposition techniques to break down code translation into sub-problems. The final code is constructed using code snippets of each query chunk, extracted from user feedback or selectively generated from a generative model. Our evaluation indicates comparable or improved performance compared to other models. Moreover, the methodology offers straightforward and interpretable approaches, which enable in-depth examination of unexpected results and facilitate insights for potential enhancements. We also illustrate that user feedback can substantially improve code translation models without re-training. Ultimately, we develop a preliminary GUI application to demonstrate the utility of our methodology in simplifying customization and assessment of suggested code for users.

PPM: Automated Generation of Diverse Programming Problems for Benchmarking Code Generation Models

PPM: Automated Generation of Diverse Programming Problems for Benchmarking Code Generation Models

The ROI of AI in lexicography

Large Language Models (LLMs) are being used for many language-based tasks, including translation, summarization and paraphrasing, sentiment analysis, and for content-generation tasks, such as code generation, answering search queries in natural language, and to power chatbots in customer service and other domains. Since much modern lexicography is based on investigation and analysis of large-scale corpora analogous to the (much larger) corpora used to train LLMs, we hypothesize that LLMs could be used for typical lexicographic tasks. A commercially-available LLM API (OpenAI’s ChatGPT gpt-3.5-turbo) was used to complete typical lexicographic tasks, such as headword expansion, phrase and form finding, and creation of definitions and examples. The results showed that the output of this LLM is not up to the standard of human editorial work, requiring significant oversight because of errors and “hallucinations” (the tendency of LLMs to invent facts). In addition, the externalities of LLM use, including concerns about environmental impact and replication of bias, add to the overall cost.

Wellness Perception and Spiritual Wellness Practices of women of the Irula Tribe of the Nilgiris Biosphere Reserve, India

Spiritual Wellness promotion is primordial at a fundamental level. For Indigenous Peoples across the world spiritual wellness is their way of life. In India we have many traditional and tribal communities that practice a holistic, nature-centric spiritual wellness. Towards learning more about Indigenous communities and their spiritual wellness conducted this exploratory study among the Irulas of the Nilgiris District in Tamil Nadu India. Being one of earlier settlers of the Nilgiris Biosphere Reserve, the Irulas have played an important role in the maintenance of the forests and environment of the Nilgiris.Methods: The study design is an ethnography qualitative study using In-depth interviews conducted using Dr. Kovach’s Conversation methodology for researching and documenting Indigenous Ontology. The Indigenous peoples are one of the Primitive Tribes of the Nilgiris District in Southern India known as the Irulas. The settlements that were accessed for the interview include villages in the eastern and western slopes of the Nilgiris. Results: Using basic qualitative analysis of coding and theme generation, data allowed to derive 4 themes about spiritual wellness namely: What is Spirituality, Importance of participation, Sustaining Practices and Conviction. There is a shift and adaptation to contemporary spiritual practices driven by lack of access to their forests and grasslands and also because of education and exposure of the younger generation.Conclusion: Development over time has resulted in changes in their natural environment as well as social environment leading to generational perception and practices of spirituality among women of the Irula tribes of the Nilgiris.

On CO2 efficiency and trade-offs between sustainability, safety, economy and resilience in infrastructure developments

The escalating crisis of climate change underscores the pressing need to reduce global CO2e emissions substantially. At the same time, a major increase is expected in global demand for infrastructure systems. In meeting this challenge, we here provide and illustrate a novel modeling framework that affords decision support in pursuit of sustainable developments of infrastructure systems. Based on this framework, for the case of roadway traffic infrastructure systems, it is shown that different strategies for the development of roadway traffic infrastructure systems, yielding approximately the same economic benefits to society, can differ in their embedded carbon footprint by up to a factor of three. This knowledge has the potential to contribute significantly to sustainable developments in general and mitigation of climate change in particular. Our models and results may readily assist decision-makers in the construction industry, at national and supra-national scales, in identifying optimal strategies for infrastructure developments. Furthermore, our work may straightforwardly support regulatory bodies and code writers in defining requirements for material and structural performances for the next generation of standards and codes.

An optimized deep network-based intermediate code generation for the mathematical expression

An optimized deep network-based intermediate code generation for the mathematical expression

Formal design, verification and implementation of robotic controller software via RoboChart and RoboTool

Current practice in simulation and implementation of robot controllers is usually undertaken with guidance from high-level design diagrams and pseudocode. Thus, no rigorous connection between the design and the development of a robot controller is established. This paper presents a framework for designing robotic controllers with support for automatic generation of executable code and automatic property checking. A state-machine based notation, RoboChart, and a tool (RoboTool) that implements the automatic generation of code and mathematical models from the designed controllers are presented. We demonstrate the application of RoboChart and its related tool through a case study of a robot performing an exploration task. The automatically generated code is platform independent and is used in both simulation and two different physical robotic platforms. Properties are formally checked against the mathematical models generated by RoboTool, and further validated in the actual simulations and physical experiments. The tool not only provides engineers with a way of designing robotic controllers formally but also paves the way for correct implementation of robotic systems.

An empirical study of prompt mode in code generation based on ChatGPT

An empirical study of prompt mode in code generation based on ChatGPT

Efficient Spin-Adapted Implementation of Multireference Algebraic Diagrammatic Construction Theory. I. Core-Ionized States and X-ray Photoelectron Spectra.

We present an efficient implementation of multireference algebraic diagrammatic construction theory (MR-ADC) for simulating core-ionized states and X-ray photoelectron spectra (XPS). Taking advantage of spin adaptation, automatic code generation, and density fitting, our implementation can perform calculations for molecules with more than 1500 molecular orbitals, incorporating static and dynamic correlation in the ground and excited electronic states. We demonstrate the capabilities of MR-ADC methods by simulating the XPS spectra of substituted ferrocene complexes and azobenzene isomers. For the ground electronic states of these molecules, the XPS spectra computed using the extended second-order MR-ADC method (MR-ADC(2)-X) are in a very good agreement with available experimental results. We further show that MR-ADC can be used as a tool for interpreting or predicting the results of time-resolved XPS measurements by simulating the core ionization spectra of azobenzene along its photoisomerization, including the XPS signatures of excited states and the minimum energy conical intersection. This work is the first in a series of publications reporting the efficient implementations of MR-ADC methods.

ChatGPT Code Detection: Techniques for Uncovering the Source of Code

In recent times, large language models (LLMs) have made significant strides in generating computer code, blurring the lines between code created by humans and code produced by artificial intelligence (AI). As these technologies evolve rapidly, it is crucial to explore how they influence code generation, especially given the risk of misuse in areas such as higher education. The present paper explores this issue by using advanced classification techniques to differentiate between code written by humans and code generated by ChatGPT, a type of LLM. We employ a new approach that combines powerful embedding features (black-box) with supervised learning algorithms including Deep Neural Networks, Random Forests, and Extreme Gradient Boosting to achieve this differentiation with an impressive accuracy of 98%. For the successful combinations, we also examine their model calibration, showing that some of the models are extremely well calibrated. Additionally, we present white-box features and an interpretable Bayes classifier to elucidate critical differences between the code sources, enhancing the explainability and transparency of our approach. Both approaches work well, but provide at most 85–88% accuracy. Tests on a small sample of untrained humans suggest that humans do not solve the task much better than random guessing. This study is crucial in understanding and mitigating the potential risks associated with using AI in code generation, particularly in the context of higher education, software development, and competitive programming.

Pengembangan E-Book Berbasis Pinterest “Pena Muhammad (Petualangan Nabi Muhammad)” Tentang Politik Rasulullah dalam Membangun Kota Madinah

This study aims to produce an innovative and effective approach to learning Islamic Religion (PAI) at the elementary level through the use of a Pinterest-based e-book entitled "PENA Muhammad (The Adventure of the Prophet Muhammad)". This research was conducted at SDN Cisempak, Wargaluyu Village, Tanjungmedar District, Sumedang Regency. This research uses a type of Design and Development (D&D) research with descriptive methods. Data were collected through direct observation and instrument validation. Researchers utilized Canva, Pinterest, Wordwall, Heyzine Flipbook, YouTube, and QR code generator applications and sites. The evaluation results showed a positive response from students to the "PENA Muhammad" e-book. Most students felt happy and successfully understood the material delivered through the e-book and demonstrated its effectiveness in improving their understanding of Islamic religious and political values, as well as its relevance in the context of modern life. The development of this e-book can be an effective innovation in learning PAI at the elementary level. By using an innovative approach through the Pinterest-based "PENA Muhammad" e-book, PAI learning for grade 3 elementary school students becomes more interesting and interactive. the development of this e-book can help students understand the Prophet's political material in building the City of Medina better. The use of multimedia and digital technology in learning provides an enjoyable and in-depth learning experience for students, so that they are able to understand the material better.

Contribution of code generation in order to save process costs in software development

Contribution of code generation in order to save process costs in software development

Examining Large Language Models' Code Generation Abilities: Utilizing Deep Learning for Employee Attrition Prediction

Examining Large Language Models' Code Generation Abilities: Utilizing Deep Learning for Employee Attrition Prediction

The MATERIAL framework: Modeling and AuTomatic code Generation of Edge Real-TIme AppLications under the QNX RTOS

Modern edge real-time automotive applications are becoming more complex, dynamic, and distributed, moving away from conventional static operating environments to support advanced driving assistance and autonomous driving functionalities. This shift necessitates formulating more complex task models to represent the evolving nature of these applications aptly. Modeling of real-time automotive systems is typically performed leveraging Architectural Languages (ALs) such as Amalthea, which are commonly used by the industry to describe the characteristics of processing platforms, operating systems, and tasks. However, these architectural languages are originally derived for classical automotive applications and need to evolve to meet the needs of next-generation applications. This paper proposes an automatic framework for the modeling and automatic code generation of dynamic automotive applications under the QNX RTOS. To this end, we extend Amalthea to describe chains of communicating tasks with multiple operating modes and to consider the QNX’s reservation-based scheduler, called APS, which allows providing temporal isolation between applications co-located on the same hardware platform. Finally, an evaluation is presented to compare different implementation alternatives under QNX that are automatically generated by our code generation framework.

BioCoder: a benchmark for bioinformatics code generation with large language models.

Pretrained large language models (LLMs) have significantly improved code generation. As these models scale up, there is an increasing need for the output to handle more intricate tasks and to be appropriately specialized to particular domains. Here, we target bioinformatics due to the amount of domain knowledge, algorithms, and data operations this discipline requires. We present BioCoder, a benchmark developed to evaluate LLMs in generating bioinformatics-specific code. BioCoder spans much of the field, covering cross-file dependencies, class declarations, and global variables. It incorporates 1026 Python functions and 1243 Java methods extracted from GitHub, along with 253 examples from the Rosalind Project, all pertaining to bioinformatics. Using topic modeling, we show that the overall coverage of the included code is representative of the full spectrum of bioinformatics calculations. BioCoder incorporates a fuzz-testing framework for evaluation. We have applied it to evaluate various models including InCoder, CodeGen, CodeGen2, SantaCoder, StarCoder, StarCoder+, InstructCodeT5+, GPT-3.5, and GPT-4. Furthermore, we fine-tuned one model (StarCoder), demonstrating that our training dataset can enhance the performance on our testing benchmark (by >15% in terms of Pass@K under certain prompt configurations and always >3%). The results highlight two key aspects of successful models: (i) Successful models accommodate a long prompt (>2600 tokens) with full context, including functional dependencies. (ii) They contain domain-specific knowledge of bioinformatics, beyond just general coding capability. This is evident from the performance gain of GPT-3.5/4 compared to the smaller models on our benchmark (50% versus up to 25%). All datasets, benchmark, Docker images, and scripts required for testing are available at: https://github.com/gersteinlab/biocoder and https://biocoder-benchmark.github.io/.