Data-driven benchmarks have led to significant progress in key scientific modeling domains including weather and structural biology. Here, we introduce the Zebrafish Activity Prediction Benchmark (ZAPBench) to measure progress on the problem of predicting cellular-resolution neural activity throughout an entire vertebrate brain. The benchmark is based on a novel dataset containing 4d light-sheet microscopy recordings of over 70,000 neurons in a larval zebrafish brain, along with motion stabilized and voxel-level cell segmentations of these data that facilitate development of a variety of forecasting methods. Initial results from a selection of time series and volumetric video modeling approaches achieve better performance than naive baseline methods, but also show room for further improvement. The specific brain used in the activity recording is also undergoing synaptic-level anatomical mapping, which will enable future integration of detailed structural information into forecasting methods.

We introduce a new machine learning approach for image segmentation that uses a neural network to model the conditional energy of a segmentation given an image. Our approach, combinatorial energy learning for image segmentation (CELIS) places a particular emphasis on modeling the inherent combinatorial nature of dense image segmentation problems. We propose efficient algorithms for learning deep neural networks to model the energy function, and for local optimization of this energy in the space of supervoxel agglomerations. We extensively evaluate our method on a publicly available 3-D microscopy dataset with 25 billion voxels of ground truth data. On an 11 billion voxel test set, we find that our method improves volumetric reconstruction accuracy by more than 20% as compared to two state-of-the-art baseline methods: graph-based segmentation of the output of a 3-D convolutional neural network trained to predict boundaries, as well as a random forest classifier trained to agglomerate supervoxels that were generated by a 3-D convolutional neural network.

An agglomerative clustering algorithm merges the most similar pair of clusters at every iteration. The function that evaluates similarity is traditionally hand- designed, but there has been recent interest in supervised or semisupervised settings in which ground-truth clustered data is available for training. Here we show how to train a similarity function by regarding it as the action-value function of a reinforcement learning problem. We apply this general method to segment images by clustering superpixels, an application that we call Learning to Agglomerate Superpixel Hierarchies (LASH). When applied to a challenging dataset of brain images from serial electron microscopy, LASH dramatically improved segmentation accuracy when clustering supervoxels generated by state of the boundary detection algorithms. The naive strategy of directly training only supervoxel similarities and applying single linkage clustering produced less improvement.

Recent studies have shown that machine learning can improve the accuracy of detecting object boundaries in images. In the standard approach, a boundary detector is trained by minimizing its pixel-level disagreement with human boundary tracings. This naive metric is problematic because it is overly sensitive to boundary locations. This problem is solved by metrics provided with the Berkeley Segmentation Dataset, but these can be insensitive to topological differences, such as gaps in boundaries. Furthermore, the Berkeley metrics have not been useful as cost functions for supervised learning. Using concepts from digital topology, we propose a new metric called the warping error that tolerates disagreements over boundary location, penalizes topological disagreements, and can be used directly as a cost function for learning boundary detection, in a method that we call Boundary Learning by Optimization with Topological Constraints (BLOTC). We trained boundary detectors on electron microscopic images of neurons, using both BLOTC and standard training. BLOTC produced substantially better performance on a 1.2 million pixel test set, as measured by both the warping error and the Rand index evaluated on segmentations generated from the boundary labelings. We also find our approach yields significantly better segmentation performance than either gPb-OWT-UCM or multiscale normalized cut, as well as Boosted Edge Learning trained directly on our data.

Many image segmentation algorithms first generate an affinity graph and then partition it. We present a machine learning approach to computing an affinity graph using a convolutional network (CN) trained using ground truth provided by human experts. The CN affinity graph can be paired with any standard partitioning algorithm and improves segmentation accuracy significantly compared to standard hand-designed affinity functions. We apply our algorithm to the challenging 3D segmentation problem of reconstructing neuronal processes from volumetric electron microscopy (EM) and show that we are able to learn a good affinity graph directly from the raw EM images. Further, we show that our affinity graph improves the segmentation accuracy of both simple and sophisticated graph partitioning algorithms. In contrast to previous work, we do not rely on prior knowledge in the form of hand-designed image features or image preprocessing. Thus, we expect our algorithm to generalize effectively to arbitrary image types.

We present an approach to low-level vision that combines two main ideas: the use of convolutional networks as an image processing architecture and an unsupervised learning procedure that synthesizes training samples from specific noise models. We demonstrate this approach on the challenging problem of natural image denoising. Using a test set with a hundred natural images, we find that convolutional networks provide comparable and in some cases superior performance to state of the art wavelet and Markov random field (MRF) methods. Moreover, we find that a convolutional network offers similar performance in the blind denoising setting as compared to other techniques in the non-blind setting. We also show how convolutional networks are mathematically related to MRF approaches by presenting a mean field theory for an MRF specially designed for image denoising. Although these approaches are related, convolutional networks avoid computational difficulties in MRF approaches that arise from probabilistic learning and inference. This makes it possible to learn image processing architectures that have a high degree of representational power (we train models with over 15,000 parameters), but whose computational expense is significantly less than that associated with inference in MRF approaches with even hundreds of parameters.

Convolutional networks have achieved a great deal of success in high-level vision problems such as object recognition. Here we show that they can also be used as a general method for low-level image processing. As an example of our approach, convolutional networks are trained using gradient learning to solve the problem of restoring noisy or degraded images. For our training data, we have used electron microscopic images of neural circuitry with ground truth restorations provided by human experts. On this dataset, Markov random field (MRF), conditional random field (CRF), and anisotropic diffusion algorithms perform about the same as simple thresholding, but superior performance is obtained with a convolutional network containing over 34,000 adjustable parameters. When restored by this convolutional network, the images are clean enough to be used for segmentation, whereas the other approaches fail in this respect. We do not believe that convolutional networks are fundamentally superior to MRFs as a representation for image processing algorithms. On the contrary, the two approaches are closely related. But in practice, it is possible to train complex convolutional networks, while even simple MRF models are hindered by problems with Bayesian learning and inference procedures. Our results suggest that high model complexity is the single most important factor for good performance, and this is possible with convolutional networks.

There is little consensus about the computational function of top-down synaptic connections in the visual system. Here we explore the hypothesis that top-down connections, like bottom-up connections, reflect partwhole relationships. We analyze a recurrent network with bidirectional synaptic interactions between a layer of neurons representing parts and a layer of neurons representing wholes. Within each layer, there is lateral inhibition. When the network detects a whole, it can rigorously enforce part-whole relationships by ignoring parts that do not belong. The network can complete the whole by filling in missing parts. The network can refuse to recognize a whole, if the activated parts do not conform to a stored part-whole relationship. Parameter regimes in which these behaviors happen are identified using the theory of permitted and forbidden sets [3, 4]. The network behaviors are illustrated by recreating Rumelhart and McClelland's "interactive activation" model [7]. In neural network models of visual object recognition [2, 6, 8], patterns of synaptic connectivity often reflect part-whole relationships between the features that are represented by neurons. For example, the connections of Figure 1 reflect the fact that feature B both contains simpler features A1, A2, and A3, and is contained in more complex features C1, C2, and C3. Such connectivity allows neurons to follow the rule that existence of the part is evidence for existence of the whole. By combining synaptic input from multiple sources of evidence for a feature, a neuron can "decide" whether that feature is present. 1 The synapses shown in Figure 1 are purely bottom-up, directed from simple to complex features. However, there are also top-down connections in the visual system, and there is little consensus about their function. One possibility is that top-down connections also reflect part-whole relationships. They allow feature detectors to make decisions using the rule that existence of the whole is evidence for existence of its parts. In this paper, we analyze the dynamics of a recurrent network in which part-whole relationships are stored as bidirectional synaptic interactions, rather than the unidirectional interactions of Figure 1. The network has a number of interesting computational capabilities. When the network detects a whole, it can rigorously enforce part-whole relationships Synaptic connectivity may reflect other relationships besides part-whole. For example, invariances can be implemented by connecting detectors of several instances of the same feature to the same target, which is consequently an invariant detector of the feature. 1

Many problems in voice recognition and audio processing involve feature extraction from raw waveforms. The goal of feature extraction is to reduce the dimensionality of the audio signal while preserving the informative signatures that, for example, distinguish different phonemes in speech or identify particular instruments in music. If the acoustic variability of a data set is described by a small number of continuous features, then we can imagine the data as lying on a low dimensional manifold in the high dimensional space of all possible waveforms. Locally linear embedding (LLE) is an unsupervised learning algorithm for feature extraction in this setting. In this paper, we present results from the exploratory analysis and visualization of speech and music by LLE.