Modern businesses need to process vast amounts of transactions quickly and accurately. Online Transaction Processing (OLTP) systems are designed to handle large numbers of simple, quick transactions such as online banking, order entry, and retail sales. However, traditional OLTP systems often face write contention, which occurs when multiple transactions attempt to modify the same data… →

Bilevel optimization (BO) is a growing field of research, gaining attention for its success in various machine learning tasks like hyperparameter optimization, meta-learning, and reinforcement learning. BO involves a two-level structure where the solution to the outer problem depends on the solution to the inner problem. However, BO is not widely used for large-scale problems,… →

Business data analysis is a field that focuses on extracting actionable insights from extensive datasets, crucial for informed decision-making and maintaining a competitive edge. Traditional rule-based systems, while precise, need help with the complexity and dynamism of modern business data. On the other hand, Artificial Intelligence (AI) models, particularly Large Language Models (LLMs), excel in… →

Ensuring the safety and moderation of user interactions with modern Language Models (LLMs) is a crucial challenge in AI. These models, if not properly safeguarded, can produce harmful content, fall victim to adversarial prompts (jailbreaks), and inadequately refuse inappropriate requests. Effective moderation tools are necessary to identify malicious intent, detect safety risks, and evaluate the… →

It is observed that LLMs often struggle to retrieve relevant information from the middle of long input contexts, exhibiting a “lost-in-the-middle” behavior. The research paper addresses the critical issue of the performance of large language models (LLMs) when handling longer-context inputs. Specifically, LLMs like GPT-3.5 Turbo and Mistral 7B often struggle with accurately retrieving information… →

Concept-based learning (CBL) in machine learning emphasizes using high-level concepts from raw features for predictions, enhancing model interpretability and efficiency. A prominent type, the concept-based bottleneck model (CBM), compresses input features into a low-dimensional space to capture essential data while discarding non-essential information. This process enhances explainability in tasks like image and speech recognition. However,… →

Large language models (LLMs) have gained significant attention in recent years, but ensuring their safe and ethical use remains a critical challenge. Researchers are focused on developing effective alignment procedures to calibrate these models to adhere to human values and safely follow human intentions. The primary goal is to prevent LLMs from engaging in unsafe… →