The discipline of computational mathematics continuously seeks methods to bolster the reasoning capabilities of large language models (LLMs). These models play a pivotal role in diverse applications ranging from data analysis to artificial intelligence, where precision in mathematical problem-solving is crucial. Enhancing these models’ ability to handle complex calculations and reasoning autonomously is paramount to… →

Integrating visual and textual data in artificial intelligence forms a crucial nexus for developing systems like human perception. As AI continues to evolve, seamlessly combining these data types is advantageous and essential for creating more intuitive and effective technologies. The primary challenge confronting this sector is the need for models to efficiently and accurately process… →

Despite their significant contributions to deep learning, LSTMs have limitations, notably in revising stored information. For instance, when faced with the Nearest Neighbor Search problem, where a sequence needs to find the most similar vector, LSTMs struggle to update stored values when encountering a closer match later in the sequence. This inability to revise storage… →

Cross-encoder (CE) models evaluate similarity by simultaneously encoding a query-item pair, outperforming the dot-product with embedding-based models at estimating query-item relevance. Current methods perform k-NN search with CE by approximating the CE similarity with a vector embedding space fit with dual-encoders (DE) or CUR matrix factorization. However, DE-based methods face challenges from poor recall because… →

Transformers have taken the machine learning world by storm with their powerful self-attention mechanism, achieving state-of-the-art results in areas like natural language processing and computer vision. However, when it came to graph data, which is ubiquitous in domains such as social networks, biology, and chemistry, the classic Transformer models hit a major bottleneck due to… →

IBM has made a great advancement in the field of software development by releasing a set of open-source Granite code models designed to make coding easier for people everywhere. This action stems from the realization that, although software plays a critical role in contemporary society, the process of coding is still difficult and time-consuming. Even… →

In Natural Language Processing (NLP) tasks, data cleaning is an essential step before tokenization, particularly when working with text data that contains unusual word separations such as underscores, slashes, or other symbols in place of spaces. Since common tokenizers frequently rely on spaces to split text into distinct tokens, this problem can have a major… →