nice_mnist

→NICEModel19.2M

nice_mnist(pretrained: bool = False, overrides: object = {})

source edit

Construct the NICE flow for the MNIST setup.

Paper-faithful MNIST configuration from Dinh et al., 2014 Figure 3: 784 dimensions, four additive coupling layers whose coupling networks have five hidden layers of 1000 rectified units, an odd/even input partition, and a standard logistic prior.

Model Size

Parameters

pretrainedbool= False

Reserved for future weight registration; currently a no-op.

**overridesobject= {}

Optional NICEConfig field overrides forwarded into the underlying config.

Returns

NICEModel

Bare bijection configured with the MNIST setup and any overrides.

Notes

Reference: Dinh, Krueger, and Bengio, "NICE: Non-linear Independent Components Estimation", ICLR Workshop, 2015 (arXiv:1410.8516), Figure 3. Reported test log-likelihood: 1980.50 nats, on data dequantised with uniform 1/256 noise and rescaled to $[0, 1]^D$ (no whitening for MNIST).

Examples

>>> import lucid
>>> from lucid.models.generative.nice import nice_mnist
>>> model = nice_mnist(input_dim=64, hidden_dim=16).eval()
>>> x = lucid.rand((1, 64))
>>> h, log_det = model.encode(x)
>>> h.shape, log_det.shape
((1, 64), (1,))

Used by 1

lucid.models.generative.nice

>>> import lucid >>> from lucid.models.generative.nice import nice_mnist >>> model = nice_mnist(input_dim=64, hidden_dim=16).eval() >>> x = lucid.rand((1, 64)) >>> h, log_det = model.encode(x) >>> h.shape, log_det.shape ((1, 64), (1,))