Moderate difficulty suitable for most players
Neural Architectures and Algorithmic Enigmas
Challenging puzzle for experienced crossword solvers
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Across
- 1. Technique to stabilize inputs in hidden layers
- 2. Iterative plunge down the slope of loss function
- 5. Operation sliding filters to detect spatial features
- 7. Vectorial representations capturing semantic nuance
- 10. Component compressing input into latent representation
- 11. Spatial result after applying convolution filters
- 14. Function converting logits into probability distribution
Down
- 3. Error signal’s path traced backwards to adjust weights
- 4. Technique reigning in complexity to prevent overfit
- 6. Implicit feature trick binding inputs through functions
- 8. Learning guided by labelled examples, explicit feedback
- 9. Network reconstructing input via bottleneck layers
- 12. Memory-empowered recurrent unit handling sequences
- 13. Random neuron omission to foster robust learning
- 15. Algorithm steering weight updates towards minima