Multimodal large language models (MLLMs) showed impressive results in various vision-language tasks by combining advanced auto-regressive language models with visual encoders. These models generated responses using visual and text inputs, with visual features from an image encoder processed before the text embeddings. However, there remains a big gap in understanding the inner mechanisms behind how… →
Ensuring AI models provide faithful and reliable explanations of their decision-making processes is still challenging. Faithfulness in the sense of explanations faithfully representing the underlying logic of a model prevents false confidence in AI systems, which is critical for healthcare, finance, and policymaking. Existing paradigms for interpretability—intrinsic (focused on inherently interpretable models) and post-hoc (providing… →
In the age of data-driven decision-making, access to high-quality and diverse datasets is crucial for training reliable machine learning models. However, acquiring such data often comes with numerous challenges, ranging from privacy concerns to the scarcity of domain-specific labeled samples. Traditional data collection and annotation processes are resource-intensive, slow, and may suffer from bias or… →
Integration of Reinforcement Learning RL with large language models catalyzes LLM’s performance on distinct specialty tasks such as robotics control or natural language processing that require sequential decision-making. Offline RL is one such technique in the spotlight today that works with static datasets without additional engagements. However, despite its utility in single-turn scenarios, Offline RL… →
Recommender systems (RS) are essential for generating personalized suggestions based on user preferences, historical interactions, and item attributes. These systems enhance user experience by helping individuals discover relevant content, such as movies, music, books, or products tailored to their interests. Popular platforms like Netflix, Amazon, and YouTube leverage RS to deliver high-quality recommendations that improve… →
In the quest to develop new drugs, the journey from laboratory research to clinical application is complex and expensive. The drug discovery process involves multiple stages, including target identification, drug screening, lead optimization, and clinical trials. Each stage requires a substantial investment of time and resources, leading to a high risk of failure. More specifically,… →
Mesh generation is an essential tool with applications in various fields, such as computer graphics and animation, computer-aided design (CAD), and virtual and augmented reality. Scaling mesh generation for converting simplified images into higher-resolution ones requires substantial computational power and memory. Additionally, maintaining intricate details while managing computational resources is challenging. Specifically, models with more… →
Mistral AI, a prominent European artificial intelligence startup, has made significant strides in AI with innovative developments and strategic expansions. Founded in April 2023 by former AI researchers from Meta Platforms and Google DeepMind, Mistral AI has rapidly positioned itself as a competitor to established entities like OpenAI. In November 2024, Mistral AI announced its… →
Graph Neural Networks (GNNs) are a rapidly advancing field in machine learning, specifically designed to analyze graph-structured data representing entities and their relationships. These networks have been widely used in social network analysis, recommendation systems, and molecular data interpretation applications. A subset of GNNs, Attention-based Graph Neural Networks (AT-GNNs), employs attention mechanisms to improve predictive… →
Networking architectures are the backbone of global communication, facilitating data exchange across vast and complex infrastructures. These systems must meet the growing demands for speed, scalability, and security while ensuring seamless integration of legacy systems with innovative technologies. They also face the critical task of adapting to diverse and unstable network environments, which has become… →