Cognition and Computation
summary
Argues that convolutional neural nets map onto visual cortex motifs and uses CNNs plus RNNs to build retail features: joint image–text embeddings for catalog search, an RNN that outputs multi-item visual tags, and generative texture models to synthesize prints. Targets practical constraints: typical merchant catalogs contain ~5,000 SKUs and ~20,000 images (platforms may host ~30,000 items), so models use fine‑tuning rather than training from scratch. Mentions neuromorphic work (TrueNorth) and that the brain runs at ~25 W, pushing energy‑efficient inference and smaller model footprints.
AI-written summary · about
chapters (12)
- 0:00Start
- 0:24What Is Cognitive Computing
- 0:42Embodied Intelligence
- 4:35Spiking Code
- 7:25Why Is Vision Difficult
- 14:59Artificial Neural Network Neurons
- 17:44Wiring Diagram of the Human of the Primate Brain
- 19:44What Is the Energy Consumption of the Brain
- 23:59The Visual Browse Use Case
- 34:31Use of Cognitive Print
- 36:19Build a Conversational System
- 48:01Learning with Clues
related talks
42mDeep Learning for Computer Vision
34mComputer Vision and NLP for Multi-Task Fashion Attrb. Modeling
37mCode First, Math Later: Learning Neural Nets Through Implementation and Examples
01h 14mTissue vs silicone musings on the future of deep learning hardware and software
33mMachine Learning in Visual Search
28mLarge Multimodal Models: Understanding the State of the Art in Compute
04mUsing Computer Vision to Understand Big Visual Data
03mRethinking software engineering in the AI era