From Classification to Conversation - by Alfredo Canziani | скачать в хорошем качестве

From Classification to Conversation - by Alfredo Canziani |

From Classification to Conversation: The evolution of Neural Networks to Transformers - A Talk By Alfredo Canziani, Assistant Professor of Computer Science @NYU 🍭 "This talk traces the transformative journey of neural network architectures that led to today's powerful language models. We begin by visualising how basic neural networks warp data spaces to classify non-linearly separable patterns This is the fundamental building block of modern AI. We then explore how recurrent neural networks (RNNs) extended these capabilities to process variable-length sequences through recursion, opening doors to language processing. The presentation culminates with the breakthrough attention mechanism that addressed critical RNN limitations, and how the transformer architecture's innovative design eliminated the final constraints, enabling the massive parallelisation that powers today's large language models. Through visual demonstrations and clear explanations, you will gain insight into the architectural innovations that drove AI's remarkable evolution from simple classification to sophisticated language understanding." Watch Afredo’s talk at Py4AI 2025, the international conference where Python and AI enthusiasts connect, learn, and explore the latest advancements in the field! May 10, 2025 | Pavia, Italy #py4ai #techconference #artificialintelligence #techtalks #AlfredoCanziani #NYU #ai