Metric learning seeks a transformation of the feature space that enhances prediction quality for a given task. In this work we provide PAC-style sample complexity rates for supervised metric learning. We give matching lower- and upper-bounds showing that sample complexity scales with the representation dimension when no assumptions are made about the underlying data distribution. In addition, by leveraging the structure of the data distribution, we provide rates fine-tuned to a specific notion of the intrinsic complexity of a given dataset, allowing us to relax the dependence on representation dimension. We show both theoretically and empirically that augmenting the metric learning optimization criterion with a simple norm-based regularization is important and can help adapt to a dataset’s intrinsic complexity yielding better generalization, thus partly explaining the empirical success of similar regularizations reported in previous works.

The generation of four-dimensional (4D) confocal datasets; consisting of 3D image sequences over time; provides an excellent methodology to capture cellular behaviors involved in developmental processes.  The ability to track and follow cell movements is limited by sample movements that occur due to drift of the sample or, in some cases, growth during image acquisition. Tracking cells in datasets affected by drift and/or growth will incorporate these movements into any analysis of cell position. This may result in the apparent movement of static structures within the sample. Therefore prior to cell tracking, any sample drift should be corrected. Using the open source Fiji distribution (1)  of ImageJ (2,3) and the incorporated LOCI tools (4), we developed the Correct 3D drift plug-in to remove erroneous sample movement in confocal datasets. This protocol effectively compensates for sample translation or alterations in focal position by utilizing phase correlation to register each time-point of a four-dimensional confocal datasets while maintaining the ability to visualize and measure cell movements over extended time-lapse experiments.

We consider the problem of estimating the parameters of a linear autoregressive model with sub-Gaussian innovations from a limited sequence of consecutive observations. Assuming that the parameters are compressible, we analyze the performance of the ℓ1-regularized least squares as well as a greedy estimator of the parameters and characterize the sampling trade-offs required for stable recovery in the non-asymptotic regime. Our results extend those of compressed sensing for linear models where the covariates are i.i.d. and independent of the observation history to autoregressive processes with highly inter-dependent covariates. We also derive sufficient conditions on the sparsity level that guarantee the minimax optimality of the ℓ1-regularized least squares estimate. Applying these techniques to simulated data as well as real-world datasets from crude oil prices and traffic speed data confirm our predicted theoretical performance gains in terms of estimation accuracy and model selection.

The nociceptive flexor withdrawal reflex has an august place in the history of neuroscience. In this issue of Neuron, Hilde et al. (2016) advance our understanding of this reflex by characterizing the molecular identity and circuit connectivity of component interneurons. They assess how a DNA-binding factor Satb2 controls cell position, molecular identity, pre-and postsynaptic targeting, and function of a population of inhibitory sensory relay interneurons that serve to integrate both proprioceptive and nociceptive afferent information.

The nociceptive flexor withdrawal reflex has an august place in the history of neuroscience. In this issue of Neuron, Hilde et al. (2016) advance our understanding of this reflex by characterizing the molecular identity and circuit connectivity of component interneurons. They assess how a DNA-binding factor Satb2 controls cell position, molecular identity, pre-and postsynaptic targeting, and function of a population of inhibitory sensory relay interneurons that serve to integrate both proprioceptive and nociceptive afferent information.

It is well known that light-sheet illumination can enable optically sectioned wide-field imaging of macroscopic samples. However, the optical sectioning capacity of a light-sheet macroscope is undermined by sample-induced scattering or aberrations that broaden the thickness of the sheet illumination. We present a technique to enhance the optical sectioning capacity of a scanning light-sheet microscope by out-of-focus background rejection. The technique, called HiLo microscopy, makes use of two images sequentially acquired with uniform and structured sheet illumination. An optically sectioned image is then synthesized by fusing high and low spatial frequency information from both images. The benefits of combining light-sheet macroscopy and HiLo background rejection are demonstrated in optically cleared whole mouse brain samples, using both green fluorescent protein (GFP)-fluorescence and dark-field scattered light contrast.

Neural circuits, governed by a complex interplay between excitatory and inhibitory neurons, are the substrate for information processing, and the organization of synaptic connectivity in neural network is an important determinant of circuit function. Here, we analyzed the fine structure of connectivity in hippocampal CA1 excitatory and inhibitory neurons innervated by Schaffer collaterals (SCs) using mGRASP in male mice. Our previous study revealed spatially structured synaptic connectivity between CA3-CA1 pyramidal cells (PCs). Surprisingly, parvalbumin-positive interneurons (PVs) showed a significantly more random pattern spatial structure. Notably, application of Peters' Rule for synapse prediction by random overlap between axons and dendrites enhanced structured connectivity in PCs, but, by contrast, made the connectivity pattern in PVs more random. In addition, PCs in a deep sublayer of striatum pyramidale appeared more highly structured than PCs in superficial layers, and little or no sublayer specificity was found in PVs. Our results show that CA1 excitatory PCs and inhibitory PVs innervated by the same SC inputs follow different connectivity rules. The different organizations of fine scale structured connectivity in hippocampal excitatory and inhibitory neurons provide important insights into the development and functions of neural networks.Understanding how neural circuits generate behavior is one of the central goals of neuroscience. An important component of this endeavor is the mapping of fine-scale connection patterns that underlie, and help us infer, signal processing in the brain. Here, using our recently developed synapse detection technology (mGRASP and neuTube), we provide detailed profiles of synaptic connectivity in excitatory (CA1 pyramidal) and inhibitory (CA1 parvalbumin-positive) neurons innervated by the same presynaptic inputs (CA3 Schaffer collaterals). Our results reveal that these two types of CA1 neurons follow different connectivity patterns. Our new evidence for differently structured connectivity at a fine scale in hippocampal excitatory and inhibitory neurons provides a better understanding of hippocampal networks and will guide theoretical and experimental studies.

Many scientific software platforms provide plugin mechanisms that simplify the integration, deployment, and execution of externally developed functionality. One of the most widely used platforms in the imaging space is Fiji, a popular open-source application for scientific image analysis. Fiji incorporates and builds on the ImageJ and ImageJ2 platforms, which provide a powerful plugin architecture used by thousands of plugins to solve a wide variety of problems. This capability is a major part of Fiji's success, and it has become a widely used biological image analysis tool and a target for new functionality. However, a plugin-based software architecture cannot unify disparate platforms operating on incompatible data structures; interoperability necessitates the creation of adaptation or "bridge" layers to translate data and invoke functionality. As a result, while platforms like Fiji enable a high degree of interconnectivity and extensibility, they were not fundamentally designed to integrate across the many data types, programming languages, and architectural differences of various software help address this challenge, we present SciJava Ops, a foundational software library for expressing algorithms as plugins in a unified and extensible way. Continuing the evolution of Fiji's SciJava plugin mechanism, SciJava Ops enables users to harness algorithms from various software platforms within a central execution environment. In addition, SciJava Ops automatically adapts data into the most appropriate structure for each algorithm, allowing users to freely and transparently combine algorithms from otherwise incompatible tools. While SciJava Ops is initially distributed as a Fiji update site, the framework does not require Fiji, ImageJ, or ImageJ2, and would be suitable for integration with additional image analysis platforms.

Automatic alignment (registration) of 3D images of adult fruit fly brains is often influenced by the significant displacement of the relative locations of the two optic lobes (OLs) and the center brain (CB). In one of our ongoing efforts to produce a better image alignment pipeline of adult fruit fly brains, we consider separating CB and OLs and align them independently. This paper reports our automatic method to segregate CB and OLs, in particular under conditions where the signal to noise ratio (SNR) is low, the variation of the image intensity is big, and the relative displacement of OLs and CB is substantial. We design an algorithm to find a minimum-cost 3D surface in a 3D image stack to best separate an OL (of one side, either left or right) from CB. This surface is defined as an aggregation of the respective minimum-cost curves detected in each individual 2D image slice. Each curve is defined by a list of control points that best segregate OL and CB. To obtain the locations of these control points, we derive an energy function that includes an image energy term defined by local pixel intensities and two internal energy terms that constrain the curve’s smoothness and length. Gradient descent method is used to optimize this energy function. To improve both the speed and robustness of the method, for each stack, the locations of optimized control points in a slice are taken as the initialization prior for the next slice. We have tested this approach on simulated and real 3D fly brain image stacks and demonstrated that this method can reasonably segregate OLs from CBs despite the aforementioned difficulties.