Neural networks (NNs) remarkably transform high-dimensional data into compact, lower-dimensional latent spaces. While researchers traditionally focus on model outputs like classification or generation, understanding the internal representation geometry has emerged as a critical area of investigation. These internal representations offer profound insights into neural network functionality, enabling researchers to repurpose learned features for downstream tasks…
Transformers have been the foundation of large language models (LLMs), and recently, their application has expanded to search problems in graphs, a foundational domain in computational logic, planning, and AI. Graph search is integral to solving tasks requiring systematically exploring nodes and edges to find connections or paths. Despite transformers’ apparent adaptability, their ability to…
Wireless communication is the foundation of modern systems, enabling critical applications in military, commercial, and civilian domains. Its increasing prevalence has changed daily life and operations worldwide while introducing serious security threats. Attackers exploit these vulnerabilities to intercept sensitive data, disrupt communications, or conduct targeted attacks, compromising confidentiality and functionality. While encryption is a critical…
Large language models (LLMs) have made important advances in artificial intelligence, with superior performance on various tasks as their parameters and training data grow. GPT-3, PaLM, and Llama-3.1 perform well in many applications with billions of parameters. However, when implemented in low-power platforms, scaling LLMs poses severe difficulties regarding training and inference queries. While it…
Sequential Recommendation systems have crucial applications in industries like e-commerce and streaming services. These systems collect and analyze the user interaction data over time to predict their preferences. However, the ID-based representations of users and items these systems rely on face critical drawbacks when transferring the same model to a new system. The new system…
LLMs like GPT-4 and LLaMA have gained significant attention for their exceptional capabilities in natural language inference, summarization, and question-answering tasks. However, these models often generate outputs that appear credible but include inaccuracies, fabricated details, or misleading information, a phenomenon termed hallucinations. This issue presents a critical challenge for deploying LLMs in applications where precision…
Large Vision Language Models (LVLMs) have demonstrated significant advancements across various challenging multi-modal tasks over the past few years. Their ability to interpret visual information in figures, known as visual perception, relied on visual encoders and multimodal training. Even with these advancements, visual perception errors still cause many mistakes in LVLMs and impact their ability…
Training AI models today isn’t just about designing better architectures—it’s also about managing data efficiently. Modern models require vast datasets and need those datasets delivered quickly to GPUs and other accelerators. The problem? Traditional data loading systems often lag behind, slowing everything down. These older systems rely heavily on process-based methods that struggle to keep…