Modelling a text corpus using Deep Boltzmann Machines in python
summary
Deep Boltzmann machines implemented in Python using bumpy model 10,000 abstracts as word‑count visibles with a softmax likelihood, pretrained as RBMs then fine‑tuned; architecture is two layers of 50 binary units each, contrastive divergence with K=1 for pretraining and dropout (1-in-10) during learning. Outputs embed documents into 50D then t‑SNE + k‑means finds clusters whose aggregated activation maps align to specific brain regions (motor, amygdala), showing the pipeline runs end‑to‑end and produces meaningful groupings; code is on GitHub.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:09Text mining in neuroscience
- 8:49Boltzmann machines overview
- 10:33Restricted Boltzmann machines
- 14:05Training RBMs in Python
- 17:16Overfitting and dropout
- 18:19RBMs for text data
- 20:45Deep Boltzmann structure
- 21:49Training with variational inference
- 23:34Pre-training and fine-tuning
- 24:16Results on neuroscience abstracts
- 27:48Visualizing embeddings
- 31:02Q&A
related talks
34mModelling a text corpus using Deep Boltzmann Machines
57mTutorial on Deep Learning III
31mTaking AI/ML to the next level: Text!
45mMachine Learning, Deep Networks and Interpretability in Systems, Cognitive and...
01h 40mA Deep Dive into NLP with PyTorch
26mBrainDrain Using Machine Learning and Brain Waves to Detect Errors in Human
25mGenerative Text Modelling: Scratching the surface
25mBig Brains, Small Targets: Whole-Brain Image Analysis with Python