The demand for intelligent and efficient digital assistants proliferates in the modern digital age. These assistants are vital for numerous tasks, including communication, learning, research, and entertainment. However, one of the primary challenges users face worldwide is finding digital assistants that can understand and interact effectively in multiple languages. Bilingual or multilingual capabilities are more…
Large-scale pre-trained vision-language models, exemplified by CLIP (Radford et al., 2021), exhibit remarkable generalizability across diverse visual domains and real-world tasks. However, their zero-shot in-distribution (ID) performance faces limitations on certain downstream datasets. Additionally, when evaluated in a closed-set manner, these models often struggle with out-of-distribution (OOD) samples from novel classes, posing safety risks in…
Numerous challenges underlying human-robot interaction exist. One such challenge is enabling robots to display human-like expressive behaviors. Traditional rule-based methods need more scalability in new social contexts, while the need for extensive, specific datasets limits data-driven approaches. This limitation becomes pronounced as the variety of social interactions a robot might encounter increases, creating a demand…
Recent advancements in machine learning and artificial intelligence (ML) techniques are used in all fields. These advanced AI systems have been made possible due to advances in computing power, access to vast amounts of data, and improvements in machine learning techniques. LLMs, which require huge amounts of data, generate human-like language for many applications. A…
In the world of advanced AI, a common challenge developers face is the security and privacy of data, especially when using external services. Many businesses and individuals have strict rules about where their sensitive information can be stored and processed. The existing solutions often involve sending data to external servers, raising concerns about compliance with…
Large language models (LLMs) like GPT-4 require substantial computational power and memory, posing challenges for their efficient deployment. While sparsification methods have been developed to mitigate these resource demands, they often introduce new complexities. For example, these techniques may require extra data structures to support the sparse representations, complicating the system architecture. The potential speedups…
A groundbreaking development is emerging in artificial intelligence and machine learning: intelligent agents that can seamlessly adapt and evolve by integrating past experiences into new and diverse tasks. These agents, central to advancing AI technology, are being engineered to perform tasks efficiently and learn and improve continuously, thereby enhancing their adaptability across various scenarios. One…
With its structured format, Tabular data dominates the data analysis landscape across various sectors such as industry, healthcare, and academia. Despite the surge in the use of images and texts for machine learning, tabular data’s inherent simplicity and interpretability have kept it at the forefront of analytical methods. However, while effective, the traditional and deep…
With the rise of Large Language Models (LLMs) in recent years, generative AI has made significant strides in the field of language processing, showcasing impressive abilities in a wide array of tasks. Given their potential in solving complex tasks, researchers have made quite a number of attempts to apply these models in the field of…
For LLMs, auto-regressive decoding is now considered the gold standard. Because LLMs generate output tokens individually, the procedure is time-consuming and expensive. Methods based on speculative sampling provide an answer to this problem. In the first, called the “draft” phase, LLMs are hypothesized at little cost; in the second, called the “verification” phase, all of…
