ARotting's picture
Publish Held-out few-shot function-inference tasks
f771f94 verified
|
Raw
History Blame Contribute Delete
1.02 kB
---
title: Neural Process Pocket
emoji: 🌊
colorFrom: violet
colorTo: cyan
sdk: gradio
sdk_version: "6.5.1"
app_file: app.py
pinned: false
---
# Neural Process Pocket
Neural Process Pocket is a Conditional Neural Process trained across a
distribution of sine functions. Five unordered context observations are encoded
into a task representation; a probabilistic decoder predicts the mean and
standard deviation at arbitrary target coordinates.
Evaluation covers RMSE, Gaussian negative log likelihood, and empirical 90%
interval coverage on unseen functions. A fixed-kernel RBF Gaussian Process is the
non-neural few-shot control.
## Verified local result
The 12,866-parameter CNP reached 1.093 RMSE, 1.242 Gaussian NLL, and 88.76%
coverage for nominal 90% intervals across 500 unseen five-context-point tasks.
The fixed-kernel Gaussian Process reached 1.259 RMSE, 1.328 NLL, and 73.63%
coverage.
```bash
uv run python projects/neural-process-pocket/train.py
uv run pytest tests/test_neural_process_pocket.py
```