Scaling AI for lifelong learning and reasoning requires the ability to transform raw inputs into abstract concepts that can be efficiently composed to form more complex ones. Our lab has a strong focus on few-shot learning for concept acquisition. In recent research, we have enabled large-scale foundation models to incrementally learn new language and visual concepts. Our current efforts extend to recognizing functional and relational concepts, as well as exploring how learned concepts can be composed hierarchically for high-level reasoning. These advancements are key to building AI systems that generalize efficiently and adapt continuously to new tasks.
Creativity is producing stimuli that are unfamiliar at first sight but quickly learnable from a few exposures. A Creator-Appraiser meta-learning loop lets a frozen diffusion model generate novel concepts the base model would not.
Published: 2026-05-15
In-Context Clustering (ICC) is a flexible LLM-based procedure for clustering data from diverse distributions.
Published: 2025-10-09
Context Tuning directly optimizes an LLM's memory representation for efficient adaptation without updating model weights.
Published: 2026-07-03
Discrete-JEPA extends the latent predictive coding JEPA framework with semantic tokenization and complementary objectives for symbolic reasoning tasks.
Published: 2025-06-22
ProCreate is a simple and easy-to-implement method to improve sample diversity and creativity of diffusion-based image generative models and to prevent training data reproduction.
Published: 2024-08-05
CoLLEGe is a meta-learning framework capable of generating flexible embeddings for new concepts using a small number of example sentences or definitions.
Published: 2024-03-22
