6 listopada 2025 09:51
Jan Chorowski: Baby Dragon Hatchling (BDH) - 19 listopada, 18:00, sala 25
Serdecznie zapraszamy na wykład Jana Chorowskiego (Pathway): Baby Dragon Hatchling (BDH), a new large language model architecture, który odbędzie się 19 listopada o 18:00 w sali 25 Instytutu Informatyki UWr.
Streszczenie:
We introduce Baby Dragon Hatchling (BDH), a new large language model architecture inspired by scale-free biological networks. BDH shows how attention can emerge from local, graph-based neuron interactions rather than centralized matrix multiplications. On one hand, BDH behaves like a brain-like, decentralized system where computation lives on synapses adjusted through Hebbian learning. On the other, it offers a dual, GPU-friendly implementation that empirically matches the performance of GPT-style Transformers. This duality makes BDH both performant and interpretable—we demonstrate the emergence of monosemantic data representations. By bridging neuroscience and machine learning, BDH points toward reasoning systems that learn and evolve continuously over time.
