Is Deep-Layered Machine Learning the Catalyst for an Artificial General Intelligence Revolution? - Itamar Arel - H+ Summit @ Harvard
4 likes1,023 views
Deep machine learning is driving advances in perception, situation inference, and capturing spatiotemporal regularities that could lead to artificial general intelligence. Deep learning uses a biologically inspired hierarchical architecture similar to cortical circuits in the brain. When combined with reinforcement learning for decision making under uncertainty, deep learning is making progress on tasks like speech recognition, image recognition, and more, bringing artificial general intelligence closer to reality. We should start discussing the moral, social, and economic implications of developing artificial general intelligence.
Is Deep-Layered Machine Learning the Catalyst for an Artificial General Intelligence Revolution? - Itamar Arel - H+ Summit @ Harvard
- 1. IS DEEP-LAYERED MACHINE LEARNING THE CATALYST FOR AN ARTIFICIAL GENERAL INTELLIGENCE REVOLUTION? Itamar Arel - Machine Intelligence Lab @ The University of Tennessee http://mil.engr.utk.edu
- 2. The AGI Technological Convergence Deep machine learning Drives perception Deep-layered Decision Learning Control Situation inference Capturing spatiotemporal regularities VLSI technology Decision making subsystem Reinforcement learning based Maps situations to decisions UT Machine Intelligence Lab http://mil.engr.utk.edu
- 3. Deep Machine Learning Biologically-inspired computational intelligence framework Goal: learning to perceive/ represent the world Hypothesis: brain represents information using a hierarchical architecture of cortical circuits UT Machine Intelligence Lab http://mil.engr.utk.edu
- 4. Deep Machine Learning (cont’) 4 Single cortical circuit Spatiotemporal modeling Bottom-up and to-down signaling Higher layers capture more abstract notions UT Machine Intelligence Lab http://mil.engr.utk.edu
- 5. Decision Making under Uncertainty We know that … Learning is driven by rewards/reinforcements Intelligence “strategic thinking” Continuous process Reinforcement learning Learning from experience Reward driven Solves “Credit assignment problem” UT Machine Intelligence Lab http://mil.engr.utk.edu
- 6. Making Headway with Deep Learning 6 Speech analytics Speakerrecognition Language ID Image recognition Emotion recognition Behavior recognition UT Machine Intelligence Lab http://mil.engr.utk.edu
- 7. Closing Thoughts … 7 The pieces of the puzzle are here Progress toward AGI is being made The right time to discuss Moral implications Socioeconomic impact Regulatory policies AGI could be in the near future … UT Machine Intelligence Lab http://mil.engr.utk.edu
- 8. 8 UT Machine Intelligence Lab http://mil.engr.utk.edu