We present ilastik, an easy-to-use interactive tool that brings machine-learning-based (bio)image analysis to end users without substantial computational expertise. It contains pre-defined workflows for image segmentation, object classification, counting and tracking. Users adapt the workflows to the problem at hand by interactively providing sparse training annotations for a nonlinear classifier. ilastik can process data in up to five dimensions (3D, time and number of channels). Its computational back end runs operations on-demand wherever possible, allowing for interactive prediction on data larger than RAM. Once the classifiers are trained, ilastik workflows can be applied to new data from the command line without further user interaction. We describe all ilastik workflows in detail, including three case studies and a discussion on the expected performance. machine learning | image analysis | software

We recently presented a new method for developing generalized gradient approximation (GGA) exchange-correlation energy functionals, using a least-squares procedure involving numerical exchange-correlation potentials and experimental energetics and nuclear gradients. In this paper we use the same method to develop a new GGA functional, denoted HCTH, based on an expansion recently suggested by Becke [J. Chem. Phys. 107, 8554 (1997)]. For our extensive training set, the new functional yields improved energetics compared to both the BLYP and B3LYP functionals [Phys. Rev. A 38, 3098 (1988); Phys. Rev. B 37, 785 (1988); J. Chem. Phys. 98, 5648 (1993); J. Phys. Chem. 98, 11623 (1994)]. The geometries of these systems, together with those of a set of transition metal compounds, are shown to be an improvement over the BLYP functional, while the reaction barriers for six hydrogen abstraction reactions are comparable to those of B3LYP. These improvements are achieved without introducing any fraction of exact orbital exchange into the new functional. We have also re-optimized the functional of Becke—which does involve exact exchange—for use in self-consistent calculations.

No abstract

Background: Regularized regression methods such as principal component or partial least squares regression perform well in learning tasks on high dimensional spectral data, but cannot explicitly eliminate irrelevant features. The random forest classifier with its associated Gini feature importance, on the other hand, allows for an explicit feature elimination, but may not be optimally adapted to spectral data due to the topology of its constituent classification trees which are based on orthogonal splits in feature space.

We present a combined report on the results of three editions of the Cell Tracking Challenge, an ongoing initiative aimed at promoting the development and objective evaluation of cell tracking algorithms. With twenty-one participating algorithms and a data repository consisting of thirteen datasets of various microscopy modalities, the challenge displays today’s state of the art in the field. We analyze the results using performance measures for segmentation and tracking that rank all participating methods. We also analyze the performance of all algorithms in terms of biological measures and their practical usability. Even though some methods score high in all technical aspects, not a single one obtains fully correct solutions. We show that methods that either take prior information into account using learning strategies or analyze cells in a global spatio-temporal video context perform better than other methods under the segmentation and tracking scenarios included in the challenge.

In many machine learning tasks it is desirable that a model's prediction transforms in an equivariant way under transformations of its input. Convolutional neural networks (CNNs) implement translational equivariance by construction; for other transformations, however, they are compelled to learn the proper mapping. In this work, we develop Steerable Filter CNNs (SFCNNs) which achieve joint equivariance under translations and rotations by design. The proposed architecture employs steerable filters to efficiently compute orientation dependent responses for many orientations without suffering interpolation artifacts from filter rotation. We utilize group convolutions which guarantee an equivariant mapping. In addition, we generalize He's weight initialization scheme to filters which are defined as a linear combination of a system of atomic filters. Numerical experiments show a substantial enhancement of the sample complexity with a growing number of sampled filter orientations and confirm that the network generalizes learned patterns over orientations. The proposed approach achieves state-of-the-art on the rotated MNIST benchmark and on the ISBI 2012 2D EM segmentation challenge.

To increase efficiency of bulk heterojunctions for photovoltaic devices, the functional morphology of active layers has to be understood, requiring visualization and discrimination of materials with very similar characteristics. Here we combine high-resolution spectroscopic imaging using an analytical transmission electron microscope with nonlinear multivariate statistical analysis for classification of multispectral image data. We obtain a visual representation showing homogeneous phases of donor and acceptor, connected by a third composite phase, depending in its extent on the way the heterojunction is fabricated. For the first time we can correlate variations in nanoscale morphology determined by material contrast with measured solar cell efficiency. In particular we visualize a homogeneously blended phase, previously discussed to diminish charge separation in solar cell devices.

Szeliski et al. published an influential study in 2006 on energy minimization methods for Markov Random Fields (MRF). This study provided valuable insights in choosing the best optimization technique for certain classes of problems.While these insights remain generally useful today, the phenomenal success of random field models means that the kinds of inference problems that have to be solved changed significantly. Specifically, the models today often include higher order interactions, flexible connectivity structures, large labelspaces of different cardinalities, or learned energy tables. To reflect these changes, we provide a modernized and enlarged study. We present an empirical comparison of 32 state-ofthe-art optimization techniques on a corpus of 2,453 energy minimization instances from diverse applications in computer vision. To ensure reproducibility, we evaluate all methods in the OpenGM 2 framework and report extensive results regarding runtime and solution quality. Key insights from our study agree with the results of Szeliski et al. for the types of models they studied. However, on new and challenging types of models our findings disagree and suggest that polyhedral methods and integer programming solvers are competitive in terms of runtime and solution quality over a large range of model types.