The ability to decompose scenes in terms of abstract building blocks is crucial for general intelligence. Where those basic building blocks share meaningful properties, interactions and other regularities across scenes, such decompositions can simplify reasoning and facilitate imagination of novel scenarios. In particular, representing perceptual observations in terms of entities should improve data efficiency and transfer performance on a wide range of tasks. Thus we need models capable of discovering useful decompositions of scenes by identifying units with such regularities and representing them in a common format. To address this problem, we have developed the Multi-Object Network (MONet). In this model, a VAE is trained end-to-end together with a recurrent attention network -in a purely unsupervised manner -to provide attention masks around, and reconstructions of, regions of images. We show that this model is capable of learning to decompose and represent challenging 3D scenes into semantically meaningful components, such as objects and background elements.

Neurons send signals to each other by means of sequences of action potentials (spikes). Ignoring variations in spike amplitude and shape that are probably not meaningful to a receiving cell, the information content, or entropy of the signal depends on only the timing of action potentials, and because there is no external clock, only the interspike intervals, and not the absolute spike times, are significant. Estimating spike train entropy is a difficult task, particularly with small data sets, and many methods of entropy estimation have been proposed. Here we present two related model-based methods for estimating the entropy of neural signals and compare them to existing methods. One of the methods is fast and reasonably accurate, and it converges well with short spike time records; the other is impractically time-consuming but apparently very accurate, relying on generating artificial data that are a statistical match to the experimental data. Using the slow, accurate method to generate a best-estimate entropy value, we find that the faster estimator converges to this value more closely and with smaller data sets than many existing entropy estimators.

No abstract

No abstract