… From conv1 we also collected l f , m on negative feature activations, using the threshold 1.0
… the terms with a weight:(4) loss = l Ω = λ p x l p x + λ f , m l f , m where we chose λ p x = 100.0 …

10.7554/eLife.25618.001 Ongoing brain oscillations are known to influence perception, and
to be reset by exogenous stimulations. Voluntary action is also accompanied by prominent …

Generative diffusion models have recently emerged as a leading approach for generating
high-dimensional data. In this paper, we show that the dynamics of these models exhibit a …

After the presentation of a visual stimulus, neural processing cascades from low-level
sensory areas to increasingly abstract representations in higher-level areas. It is often …

Neural decoding can be conceptualized as the problem of mapping brain responses back
to sensory stimuli via a feature space. We introduce (i) a novel experimental paradigm that …

This paper introduces Wasserstein variational inference, a new form of approximate Bayesian
inference based on optimal transport theory. Wasserstein variational inference uses a new …

This paper introduces the kernel mixture network, a new method for nonparametric estimation
of conditional probability densities using neural networks. We model arbitrarily complex …

… Given an input-target pair (y0,tL) we can define a quadratic loss function, l as … − 1 is equal
to the number of units in the subsequent layer indexed l. We can therefore describe the l − 1-th …

… We further regularize the input using an ℓ 1 loss on all components (pixel values) of I.
The ℓ 1 leads to the suppression of noise in the image, which otherwise easily occurs in this …

… The employed l 1 regularizer encourages sparse precision matrices as determined by
the regularization parameter λ. This maximization problem can be solved using established …