A major challenge in the field of natural language processing (NLP) is addressing the limitations of decoder-only Transformers. These models, which form the backbone of large language models (LLMs), suffer from significant issues such as representational collapse and over-squashing. Representational collapse occurs when different input sequences produce nearly identical representations, while over-squashing leads to a… →

The remarkable performance in different reasoning tasks has been demonstrated by several Large Language Models (LLMs), such as GPT-4, PaLM, and LLaMA. To further increase the functionality and performance of LLMs, there are more effective prompting methods and increasing the model size, both of which boost reasoning performance. The approaches are classified as follows: (i)… →

Dataset distillation is an innovative approach that addresses the challenges posed by the ever-growing size of datasets in machine learning. This technique focuses on creating a compact, synthetic dataset that encapsulates the essential information of a larger dataset, enabling efficient and effective model training. Despite its promise, the intricacies of how distilled data retains its… →

In today’s age, learning AI is crucial as companies increasingly rely on it for efficiency, automation, and personalization, yet not everyone is an expert in the field. Salesforce offers short courses on Trailhead, covering essential AI skills to help you become the AI hero your company needs, positioning you for new opportunities and career advancement.… →

Accurately predicting antibody structures is essential for developing monoclonal antibodies, pivotal in immune responses and therapeutic applications. Antibodies have two heavy and two light chains, with the variable regions featuring six CDR loops crucial for binding to antigens. The CDRH3 loop presents the greatest challenge due to its diversity. Traditional experimental methods for determining antibody… →

Recent advancements in machine learning have been actively used to improve the domain of healthcare. Despite performing remarkably well on various tasks, these models are often unable to provide a clear understanding of how specific visual changes affect ML decisions. These AI models have shown great promise and even human capabilities in some cases, but… →

This paper explores the domain of uncertainty quantification within large language models (LLMs) to identify scenarios where uncertainty in response to queries is significant. The study encompasses both epistemic and aleatoric uncertainties. Epistemic uncertainty arises from a lack of knowledge or data about the ground truth, whereas aleatoric uncertainty stems from inherent randomness in the… →

Recent advances in artificial intelligence, primarily driven by foundation models, have enabled impressive progress. However, achieving artificial general intelligence, which involves reaching human-level performance across various tasks, remains a significant challenge. A critical missing component is a formal description of what it would take for an autonomous system to self-improve towards increasingly creative and diverse… →