This work presents a hierarchical approach to explaining poisoned artificial intelligence (AI) models. The motivation comes from the use of AI models in security and safety critical applications, for instance, the use of AI models for classification of road traffic signs in self-driving cars. Training images of traffic signs can be poisoned by adversaries to encode malicious triggers that change trained AI model prediction from a correct traffic sign to another traffic sign in a presence of such a physically realizable trigger (e.g., sticky note or Instagram filter). We address the lack of AI model explainability by (a) designing utilization measurements of trained AI models and (b) explaining how training data are encoded in AI models based on those measurements at three hierarchical levels. The three levels are defined at graph node (computation unit), subgraph, and graph representations of poisoned and clean AI models from the TrojAI Challenge.

This work addresses the problems of (a) designing utilization measurements of trained artificial intelligence (AI) models and (b) explaining how training data are encoded in AI models based on those measurements. The problems are motivated by the lack of explainability of AI models in security and safety critical applications, such as the use of AI models for classification of traffic signs in self-driving cars. We approach the problems by introducing theoretical underpinnings of AI model utilization measurement and understanding patterns in utilization-based class encodings of traffic signs at the level of computation graphs (AI models), subgraphs, and graph nodes. Conceptually, utilization is defined at each graph node (computation unit) of an AI model based on the number and distribution of unique outputs in the space of all possible outputs (tensor-states). In this work, utilization measurements are extracted from AI models, which include poisoned and clean AI models. In contrast to clean AI models, the poisoned AI models were trained with traffic sign images containing systematic, physically realizable, traffic sign modifications (i.e., triggers) to change a correct class label to another label in a presence of such a trigger. We analyze class encodings of such clean and poisoned AI models, and conclude with implications for trojan injection and detection.

In living systems, it is frequently stated that form follows function by virtue of evolutionary pressures on organism development, but in the study of how functions emerge at the cellular level, function often follows form. We study this chicken versus egg problem of emergent structure-property relationships in living systems in the context of primary human bone marrow stromal cells cultured in a variety of microenvironments that have been shown to cause distinct patterns of cell function and differentiation. Through analysis of a publicly available catalog of three-dimensional (3D) cell shape data, we introduce a family of metrics to characterize the 'form' of the cell populations that emerge from a variety of diverse microenvironments. In particular, measures of form are considered that are expected to have direct significance for cell function, signaling and metabolic activity: dimensionality, polarizability and capacitance. Dimensionality was assessed by an intrinsic measure of cell shape obtained from the polarizability tensor. This tensor defines ellipsoids for arbitrary cell shapes and the thinnest dimension of these ellipsoids, P 1, defines a reference minimal scale for cells cultured in a 3D microenvironment. Polarizability governs the electric field generated by a cell, and determines the cell's ability to detect electric fields. Capacitance controls the shape dependence of the rate at which diffusing molecules contact the surface of the cell, and this has great significance for inter-cellular signaling. These results invite new approaches for designing scaffolds which explicitly direct cell dimensionality, polarizability and capacitance to guide the emergence of new cell functions derived from the acquired form.

The ZENO software tool computes material, solution,and suspension properties for a specified particle shape or molecular structure using path-integral and Monte Carlo methods. These properties include: capacitance, electric polarizability tensor, intrinsic conductivity, volume, gyrationtensor, hydrodynamic radius, intrinsic viscosity, friction coefficient, diffusion coefficient, sedimentation coefficient, and related quantities.

Many biomaterial scaffolds have been advanced to provide synthetic cell niches for tissue engineering and drug screening applications; however, current methods for comparing scaffold niches focus on cell functional outcomes or attempt to normalize materials properties between different scaffold formats. We demonstrate a three-dimensional (3D) cellular morphotyping strategy for comparing biomaterial scaffold cell niches between different biomaterial scaffold formats. Primary human bone marrow stromal cells (hBMSCs) were cultured on 8 different biomaterial scaffolds, including fibrous scaffolds, hydrogels, and porous sponges, in 10 treatment groups to compare a variety of biomaterial scaffolds and cell morphologies. A bioinformatics approach was used to determine the 3D cellular morphotype for each treatment group by using 82 shape metrics to analyze approximately 1000 cells. We found that hBMSCs cultured on planar substrates yielded planar cell morphotypes, while those cultured in 3D scaffolds had elongated or equiaxial cellular morphotypes with greater height. Multivariate analysis was effective at distinguishing mean shapes of cells in flat substrates from cells in scaffolds, as was the metric L-1-depth (the cell height along its shortest axis after aligning cells with a characteristic ellipsoid). The 3D cellular morphotyping technique enables direct comparison of cellular microenvironments between widely different types of scaffolds and design of scaffolds based on cell structure-function relationships.

The Endless Forest [1] is described by its creators as a “multiplayer online game and social screensaver”. Although it is an MMO (Massively Multiplayer Online) game, it runs on only a single server. The game is implemented as a Windows screensaver. Development of the game was funded by several artistic cultural institutions, including Musée d'Art Moderne Grand-Duc Jean, which also hosts the server. The gameplay is very abstract and artisticfor example, players may not chat directly with each other, but must attempt to communicate through deer-gestures.

This paper describes our on-going work to accelerate ZENO, a software tool based on Monte Carlo methods (MCMs), used for computing material properties at nanoscale. ZENO employs three main algorithms: (1) Walk on Spheres (WoS), (2) interior sampling, and (3) surface sampling. We have accelerated the first two algorithms. For the sake of brevity, the paper will discuss our work on the first one only as it is the most commonly used and the acceleration techniques were similar in both cases. WoS is a Brownian motion MCM for solving a class of partial differential equations (PDEs). It provides a stochastic solution to a PDE by estimating the probability that a random walk, which started at infinity, will hit the surface of the material under consideration. WoS is highly effective when the problem's geometry is additive, as this greatly reduces the number of walk steps needed to achieve accurate results. The walks start on the surface of an enclosing sphere and can make much larger jumps than in a direct simulation of Brownian motion. Our current implementation represents the molecular structure of nanomaterials as a union of possibly overlapping spheres. The core processing bottleneck in WoS is a Computational Geometry one, as the algorithm repeatedly determines the distance from query point to the material surface in each step of the random walk. In this paper, we present results from benchmarking spatial data structures, including several open-source implementations of k-D trees, for accelerating WoS algorithmically. The paper also presents results from our multicore and cluster parallel implementation to show that it exhibits linear strong scaling with the number of cores and compute nodes; this implementation delivers up to 4 orders of magnitude speedup compared to the original FORTRAN code when run on 8 nodes (each with dual 6-core Intel Xeon CPUs) with 24 threads per node.

Perlin's original noise function was defined in terms of generating random vectors at lattice points and then interpolating between them, but a simple approximation can be created by instead generating White noise values at the lattice points and then interpolating. The fewer interpolated points there are, the higher the frequencies in the noise will be. You can then generate fractal noise by taking a weighted sum of a series of different frequency noises (Perlin or other), generally with larger weights being used for the lower frequencies. White noise [1]

BACKGROUND:The goal of this survey paper is to overview cellular measurements using optical microscopy imaging followed by automated image segmentation. The cellular measurements of primary interest are taken from mammalian cells and their components. They are denoted as two- or three-dimensional (2D or 3D) image objects of biological interest. In our applications, such cellular measurements are important for understanding cell phenomena, such as cell counts, cell-scaffold interactions, cell colony growth rates, or cell pluripotency stability, as well as for establishing quality metrics for stem cell therapies. In this context, this survey paper is focused on automated segmentation as a software-based measurement leading to quantitative cellular measurements.METHODS:We define the scope of this survey and a classification schema first. Next, all found and manually filteredpublications are classified according to the main categories: (1) objects of interests (or objects to be segmented), (2) imaging modalities, (3) digital data axes, (4) segmentation algorithms, (5) segmentation evaluations, (6) computational hardware platforms used for segmentation acceleration, and (7) object (cellular) measurements. Finally, all classified papers are converted programmatically into a set of hyperlinked web pages with occurrence and co-occurrence statistics of assigned categories.RESULTS:The survey paper presents to a reader: (a) the state-of-the-art overview of published papers about automated segmentation applied to optical microscopy imaging of mammalian cells, (b) a classification of segmentation aspects in the context of cell optical imaging, (c) histogram and co-occurrence summary statistics about cellular measurements, segmentations, segmented objects, segmentation evaluations, and the use of computational platforms for accelerating segmentation execution, and (d) open research problems to pursue.CONCLUSIONS:The novel contributions of this survey paper are: (1) a new type of classification of cellular measurements and automated segmentation, (2) statistics about the published literature, and (3) a web hyperlinked interface to classification statistics of the surveyed papers at https://isg.nist.gov/deepzoomweb/resources/survey/index.html.

There is a need for tools to classify cells based on their three-dimensional (3D) shape. Cells exist in vivo in 3D, cells are frequently cultured within 3D scaffolds in vitro and 3D scaffolds are used for cell delivery in tissue engineering therapies. Recent work indicates that the physical structure of a tissue engineering scaffold can direct stem cell function by driving stem cells into morphologies that induce their differentiation. Thus, we have developed a rapid method for classifying cells based on their 3D shape. First, random lines are intersected with 3D Z-stacks of confocal images of stem cells. The intersection lengths are stored in histograms, which are then used to train a support vector machine (SVM) learning algorithm to distinguish between stem cells cultured on differentiation-inducing 3D scaffolds and those cultured on non-differentiating flat substrates. The trained SVM is able to properly classify the ?new? query cells over 80% of the time. The algorithm is easily parallelizable and we demonstrate its implementation on a commodity graphics processing unit (GPU). Use of a GPU to run the algorithm increases throughput by over 100-fold as compared to use of a CPU. The algorithm is also progressive, providing an approximate answer quickly and refining the answer over time. This allows further increase in the throughput of the algorithm by allowing the SVM classification scheme to terminate early if it becomes confident enough of the class of the cell being analyzed. These results demonstrate a rapid method for classifying stem cells based on their 3D shape that can be used by tissue engineers for identifying 3D tissue scaffold structures that drive stem cells into shapes that correlate with differentiation.

There is a need for tools to classify cells based on their 3D shape. Cells exist in vivo in 3D, cells are frequently cultured within 3D scaffolds in vitro, and 3D scaffolds are used for cell delivery in tissue engineering therapies. Recent work indicates that the physical structure of a tissue engineering scaffold can direct stem cell function by driving stem cells into morphologies that induce their differentiation. Thus, we have developed a rapid method for classifying cells based on their 3D shape. First, random lines are intersected with 3D Z-stacks of confocal images of stem cells. The intersection lengths are stored in histograms, which are then used to train a Support Vector Machine (SVM) learning algorithm to distinguish between stem cells cultured on differentiationinducing 3D scaffolds and those cultured on non-differentiating flat substrates. The trained SVM is able to properly classify the “new” query cells over 80% of the time. The algorithm is easily parallelizable and we demonstrate its implementation on a commodity Graphics Processing Unit (GPU). Use of a GPU to run the algorithm increases throughput by over 100-fold as compared to use of a CPU. The algorithm is also progressive, providing an approximate answer quickly and refining the answer over time. This allows further increase in the throughput of the algorithm by allowing the SVM classification scheme to terminate early if it becomes confident enough of the class of the cell being analyzed. These results demonstrate a rapid method for classifying stem cells based on their 3D shape that can be used by tissue engineers for identifying 3D tissue scaffold structures that drive stem cells into shapes that correlate with differentiation. PACS numbers: 87.57.N-, 87.64.mk Parallel Geometric Classification of Stem Cells by Their 3D Morphology 2

Direct numerical simulation (DNS) of turbulence is computationally intensive and typically relies on some form of parallel processing. The authors present techniques to map DNS computations to modern graphics processing units (GPUs), which are characterized by very high memory bandwidth and hundreds of SPMD (single-program-multiple-data) processors.

Biochemists often wish to compute surface areas of proteins. A variety of algorithms have been developed for this task, but they are designed for traditional single-processor architectures. The current trend in computer hardware is towards increasingly parallel architectures for which these algorithms are not well suited. We describe a parallel, stochastic algorithm for molecular surface area computation that maps well to the emerging multi-core architectures. Our algorithm is also progressive, providing a rough estimate of surface area immediately and refining this estimate as time goes on. Furthermore, the algorithm generates points on the molecular surface which can be used for point-based rendering. We demonstrate a GPU implementation of our algorithm and show that it compares favorably with several existing molecular surface computation programs, giving fast estimates of the molecular surface area with good accuracy.

Data from gyrokinetic turbulence codes are often difficult to visualize due their high dimensionality, the nontrivial geometry of the underlying grids, and the vast range of spatial scales. We present an interactive visualization framework that attempts to address these issues. Images from a nonlinear gyrokinetic simulation are presented.

Implicit representations have the potential to represent large volumes succinctly. In this paper we present a multiresolution and progressive implicit representation of scalar volumetric data using anisotropic Gaussian radial basis functions (RBFs) dened over an octree. Our representation lends itself well to progressive level-of-detail representations. Our RBF encoding algorithm based on a Maximum Likelihood Estimation (MLE) calculation is non-iterative, scales in a O(n log n) manner, and operates in a memory-friendly manner on very large datasets by processing small blocks at a time. We also present a GPU-based ray-casting algorithm for direct rendering from implicit volumes. Our GPU-based implicit volume rendering algorithm is accelerated by early-ray termination and empty-space skipping for implicit volumes and can render volumes encoded with 16 million RBFs at 1 to 3 frames/second. The octree hierarchy enables the GPU-based ray-casting algorithm to efciently traverse using location codes and is also suitable for view-dependent level-of-detail-based rendering.