Все выпуски

Предлагается диссипативная стохастическая динамическая модель эволюции языковых знаков, удовлетворяющая принципу «наименьшего действия» — одному из фундаментальных вариационных принципов природы. Модель предполагает пуассоновский характер потока рождения языковых знаков, экспоненциальное (показательное) распределение ассоциативно-семантического потенциала (АСП) знака и оперирует разностными стохастическими уравнениями специального вида для диссипативных процессов. Получаемые из модели распределения полисемии и частотно-ранговые распределения языковых знаков статистически значимо (по критерию Колмогорова–Смирнова) не отличаются от эмпирических распределений, полученных из представительных толковых и частотных словарей русского и английского языков.

We offer the dissipative stochastic dynamic model of the language sign evolution, satisfying to the principle of the least action, one of fundamental variational principles of the Nature. The model conjectures the Poisson nature of the birth flow of language signs and the exponential distribution of their associative-semantic potential (ASP). The model works with stochastic difference equations of the special type for dissipative processes. The equation for momentary polysemy distribution and frequency-rank distribution drawn from our model do not differs significantly (by Kolmogorov-Smirnov’s test) from empirical distributions, got from main Russian and English explanatory dictionaries as well as frequency dictionaries of them.

Интерпретируемость моделей глубокого обучения стала центром исследований, особенно в таких областях, как здравоохранение и финансы. Модели с «бутылочным горлышком», используемые для выявления концептов, стали перспективным подходом для достижения прозрачности и интерпретируемости за счет использования набора известных пользователю понятий в качестве промежуточного представления перед слоем предсказания. Однако ручное аннотирование понятий не затруднено из-за больших затрат времени и сил. В нашей работе мы исследуем потенциал больших языковых моделей (LLM) для создания высококачественных банков концептов и предлагаем мультимодальную метрику для оценки качества генерируемых концептов. Мы изучили три ключевых вопроса: способность LLM генерировать банки концептов, сопоставимые с существующими базами знаний, такими как ConceptNet, достаточность унимодального семантического сходства на основе текста для оценки ассоциаций концептов с метками, а также эффективность мультимодальной информации для количественной оценки качества генерации концептов по сравнению с унимодальным семантическим сходством концепт-меток. Наши результаты показывают, что мультимодальные модели превосходят унимодальные подходы в оценке сходства между понятиями и метками. Более того, сгенерированные нами концепты для наборов данных CIFAR-10 и CIFAR-100 превосходят те, что были получены из ConceptNet и базовой модели, что демонстрирует способность LLM генерировать высококачественные концепты. Возможность автоматически генерировать и оценивать высококачественные концепты позволит исследователям работать с новыми наборами данных без дополнительных усилий.

Interpretability in recent deep learning models has become an epicenter of research particularly in sensitive domains such as healthcare, and finance. Concept bottleneck models have emerged as a promising approach for achieving transparency and interpretability by leveraging a set of humanunderstandable concepts as an intermediate representation before the prediction layer. However, manual concept annotation is discouraged due to the time and effort involved. Our work explores the potential of large language models (LLMs) for generating high-quality concept banks and proposes a multimodal evaluation metric to assess the quality of generated concepts. We investigate three key research questions: the ability of LLMs to generate concept banks comparable to existing knowledge bases like ConceptNet, the sufficiency of unimodal text-based semantic similarity for evaluating concept-class label associations, and the effectiveness of multimodal information in quantifying concept generation quality compared to unimodal concept-label semantic similarity. Our findings reveal that multimodal models outperform unimodal approaches in capturing concept-class label similarity. Furthermore, our generated concepts for the CIFAR-10 and CIFAR-100 datasets surpass those obtained from ConceptNet and the baseline comparison, demonstrating the standalone capability of LLMs in generating highquality concepts. Being able to automatically generate and evaluate high-quality concepts will enable researchers to quickly adapt and iterate to a newer dataset with little to no effort before they can feed that into concept bottleneck models.

Large Language Models (LLMs) are transforming software engineering by bridging the gap between natural language and programming languages. These models have revolutionized communication within development teams and the Software Development Life Cycle (SDLC) by enabling developers to interact with code using natural language, thereby improving workflow efficiency. This survey examines the impact of LLMs across various stages of the SDLC, including requirement gathering, system design, coding, debugging, testing, and documentation. LLMs have proven to be particularly useful in automating repetitive tasks such as code generation, refactoring, and bug detection, thus reducing manual effort and accelerating the development process. The integration of LLMs into the development process offers several advantages, including the automation of error correction, enhanced collaboration, and the ability to generate high-quality, functional code based on natural language input. Additionally, LLMs assist developers in understanding and implementing complex software requirements and design patterns. This paper also discusses the evolution of LLMs from simple code completion tools to sophisticated models capable of performing high-level software engineering tasks. However, despite their benefits, there are challenges associated with LLM adoption, such as issues related to model accuracy, interpretability, and potential biases. These limitations must be addressed to ensure the reliable deployment of LLMs in production environments. The paper concludes by identifying key areas for future research, including improving the adaptability of LLMs to specific software domains, enhancing their contextual understanding, and refining their capabilities to generate semantically accurate and efficient code. This survey provides valuable insights into the evolving role of LLMs in software engineering, offering a foundation for further exploration and practical implementation.

Large Language Models (LLMs) are transforming software engineering by bridging the gap between natural language and programming languages. These models have revolutionized communication within development teams and the Software Development Life Cycle (SDLC) by enabling developers to interact with code using natural language, thereby improving workflow efficiency. This survey examines the impact of LLMs across various stages of the SDLC, including requirement gathering, system design, coding, debugging, testing, and documentation. LLMs have proven to be particularly useful in automating repetitive tasks such as code generation, refactoring, and bug detection, thus reducing manual effort and accelerating the development process. The integration of LLMs into the development process offers several advantages, including the automation of error correction, enhanced collaboration, and the ability to generate high-quality, functional code based on natural language input. Additionally, LLMs assist developers in understanding and implementing complex software requirements and design patterns. This paper also discusses the evolution of LLMs from simple code completion tools to sophisticated models capable of performing high-level software engineering tasks. However, despite their benefits, there are challenges associated with LLM adoption, such as issues related to model accuracy, interpretability, and potential biases. These limitations must be addressed to ensure the reliable deployment of LLMs in production environments. The paper concludes by identifying key areas for future research, including improving the adaptability of LLMs to specific software domains, enhancing their contextual understanding, and refining their capabilities to generate semantically accurate and efficient code. This survey provides valuable insights into the evolving role of LLMs in software engineering, offering a foundation for further exploration and practical implementation.