Large pre-trained generative transformers have demonstrated exceptional performance in various natural language generation tasks, using large training datasets to capture the logic of human language. However, adapting these models for certain applications through fine-tuning poses significant challenges. The computational efficiency of fine-tuning depends heavily on the model size, making it costly for researchers to work… →

Large Language Models (LLMs) have demonstrated great performance in Natural Language Processing (NLP) applications. However, they have high computational costs when fine-tuning them, which can lead to incorrect information being generated, i.e., hallucinations. Two viable strategies have been established to solve these problems: parameter-efficient methods such as Low-Rank Adaptation (LoRA) to minimize computing demands and… →

Machine learning has revolutionized various fields, offering powerful tools for data analysis and predictive modeling. Central to these models’ success is hyperparameter optimization (HPO), where the parameters that govern the learning process are tuned to achieve the best possible performance. HPO involves selecting hyperparameter values such as learning rates, regularization coefficients, and network architectures. These… →

Hypergraphs, which extend traditional graphs by allowing hyperedges to connect multiple nodes, offer a richer representation of complex relationships in fields like social networks, bioinformatics, and recommender systems. Despite their versatility, generating realistic hypergraphs is challenging due to their complexity and the need for effective generative models. While traditional methods focus on algorithmic generation with… →

Spiking Neural Networks (SNNs) hold significant promise in developing energy-efficient and biologically plausible artificial neural networks. However, a critical challenge is their limited ability to handle sequential tasks such as text classification and time-series forecasting. This limitation primarily stems from the lack of an effective spike-form positional encoding (PE) mechanism, which is crucial for capturing… →

Information management and retrieval systems are essential for businesses and organizations, whether for customer support, internal knowledge bases, academic research, or instructional purposes. It can be challenging to manage enormous data volumes while ensuring users can quickly locate what they need. Regarding privacy issues, language support, and ease of use, existing tools frequently need to… →

Large Language Models (LLMs) have become increasingly important in cybersecurity, particularly in their application to secure coding practices. As these AI-driven models can generate human-like text, they are now being utilized to detect and mitigate security vulnerabilities in software. The primary goal is to harness these models to enhance the security of code, which is… →