With the rapid advancements in machine learning, models have become increasingly capable of learning and making predictions in various industries. However, deploying these models in critical infrastructures presents a major challenge, as concerns about data privacy prevent unrestricted data sharing. Homomorphic encryption (HE) offers a solution by enabling computations on encrypted data, but it remains incompatible with machine learning models like convolutional neural networks (CNNs), due to their reliance on non-linear activation functions. To bridge this gap, this work proposes an optimized framework that replaces standard non-linear functions with homomorphically compatible approximations, ensuring secure computations while minimizing computational overhead. The proposed approach restructures the CNN architecture and introduces an efficient activation function approximation method to mitigate the performance trade-offs introduced by encryption. Experiments on CIFAR-10 achieve 94.4% accuracy with 2.42 s per single encrypted sample and 24,000 s per 10,000 encrypted samples, using a degree-4 polynomial and Softplus activation under CKKS, balancing accuracy and privacy.

This paper proposes a novel metric for evaluating the quality of images of the same category generated by a synthetic image generator. The proposed method assumes that the points representing synthetic images in a feature space lie on a manifold, and it aims to determine the curvatures of the manifold at those points. The lower the angle, the higher the quality of the generated images is considered to be. Let $\mathcal{F}$ be the set of data points in a D-dimensional feature space where each point represents one synthetically generated image. Let $\mathbb{M}$ be the corresponding manifold to match $\mathcal{F}$. For each data point $y\in\mathbb{R}^{D}$ on $\mathbb{M}$, a number of neighboring points are determined. A local subspace $\mathbb{S}$ is matched for y and a local subspace is also matched for each of its neighboring points. Then, a set of weighted angles between $\mathbb{S}$ and each neighboring local subspace are computed. The average of those weighted angles is used as a measure of curvature of $\mathbb{M}$ at y. Experimental results demonstrate the effectiveness of the proposed metric in evaluating the quality of synthetically generated images.

This paper introduces a novel method for creating a metric to measure curvature of a discretized manifold. For each data point xion a manifold, a subspace ${{\mathcal{S}}_i}$ is matched using a number of neighboring points of xi. A local subspace is also matched to each neighboring point of xi. Then, a set of weighted angles between ${{\mathcal{S}}_i}$ and each neighboring local subspace are computed and the minimum of those weighted angles is used as a measure of curvature of the manifold at xi. The average curvature for all data points on the manifold is used as metric for the manifold’s curvature estimation. This research also uses the proposed metric to show that each layer of a neural network maps an input manifold to a flatter manifold during the training process. It is observed that each successive block in a neural network generates a manifold with less curvature than that of the previous layer. Another observation is that convolutional layers always generate flatter manifolds. Furthermore, it is shown that this metric can be used as a robustness measure for a neural network. The method has been tested successfully using two datasets MNIST and Extended YaleB datasets.

Secondary structure elements in protein molecules refer to local sub-conformational regions stabilized by hydrogen bonding. Assigning Secondary Structure Elements is crucial in protein structure determination and function analysis. This work represents a recast of a previously developed classifier using ensemble of machine learning models. In this paper, we introduce new geometrical features to improve the accuracy, reduce training data set and process, and we develop and apply a post-processing step. The classifier is trained with 150K amino acids. We tested our classifier on a set of 20 protein structures and compared with previously developed classifier. The information from Protein Data Bank was used as a reference. The comparison shows that new method can produce assignments that are more aligned with PDB at 95.31% accuracy after applying a simple postprocessing step compared to 92.75% for the previous classifier.

This paper presents a Deep Neural network (DNN) system that uses a large set of geometric and categorical features for classification of secondary structure elements (SSEs) in the protein’s trace that consists of $C\alpha$ atoms on the backbone. A systematical approach is implemented for classification of protein SSE problem. This approach consists of two network architecture search (NAS) algorithms for selecting (1) network architecture and layer connectivity, and (2) regularization parameters. Each algorithm uses a different search space and they are used in succession to develop a DNN. The DNN system generates over 93% classification rate on average for multiple test sets without any post processing for amino acid configurations.

This paper presents a novel subspace segmentation algorithm that models protein $C\alpha$ traces of secondary structure elements (SSEs) as a union of subspaces. For each $C\alpha$, a set of general geometric features are considered. The algorithm first identifies the most relevant features for each SSE using a new matrix rank estimation technique and combinatorics. This is followed by grouping $C\alpha$ traces in a sliding-window so that each group represents a data point in a high-dimensional ambient space. Then, a lower dimensional subspace is matched for each SSE. When a group of unknown $C\alpha$ traces is presented, the algorithm determines a neighborhood around each $C\alpha$ and then uses two approaches to classify the $C\alpha$. In the first approach, the $C\alpha$ is represented as a data point in the ambient space and its distance to each subspace is calculated. In the second approach, a local subspace is matched to the $C\alpha$, and the separation of this local subspace from each SSE subspace is computed using geodesic distance on the Grassmannian manifold of the subspaces. The minimum point-to-subspace distance and minimum separation of subspaces are used to classify the $C\alpha$. This geometric and mathematical approach has been applied a large protein dataset and generated 85% classification rate without the need to train a large machine learning system.

This paper introduces a clustering algorithm, called principal coordinate clustering. It takes in a similarity matrix SWof a data matrix W and computes the singular value decomposition of SWto determine the principal coordinates to convert the clustering problem to a simpler domain. It is a relative of spectral clustering, however, principal coordinate clustering is easier to interpret, and gives a clear understanding of why it performs well. In a fashion, this gives intuition behind why spectral clustering works from a more simple, linear algebra perspective, beyond the typical explanations via graph cuts, or other techniques. Moreover, it was demonstrated through experimentation on real and synthetic data that the proposed method performs equally well on average as spectral clustering, and that the method has the ability to scale quite easily to truly large data.

This paper presents a novel technique for the segmentation of data W = [w1· · · wn] ⊂ RDdrawn from a union u = ∪Mi=1of subspaces {Si}Mi=1. First, an existing subspace segmentation algorithm is used to perform an initial data clustering {Ci}Mi=1, where Ci= {wi1· · ·wik} ⊂ W is the set of data from the ithcluster. Then, a local subspace LSiis matched for each Ciand the distance dijbetween LSiand each point wij∊ Ciis computed. A data-driven threshold η is computed and the data points (in Ci) whose distances to LSiare larger than η are eliminated since they are considered as outliers or erroneously clustered data points in Ci. The remaining data points Ci⊂ Ciare considered to be coming from the same subspace with high confidence. Then, {Ci}Mi=1are used in unsupervised way to train a convolution neural network to obtain a deep learning model, which is in turn used to re-cluster W. The system has been successfully implemented using the MNIST dataset and it improved the segmentation accuracy of a particular algorithm (EnSC-ORGEN) from 93.79% to 96.52%.

This paper provides a comprehensive analysis of skeleton decomposition used for segmentation of data W = [w1···WN] ⊂ ℝddrawn from a union U = ∪i=1MSiof linearly independent subspaces {Si}i=1Mof dimensions of {di}i=1M. Our previous work developed a generalized theoretical framework for computing similarity matrices by matrix factorization. Skeleton decomposition is a special case of this general theory. First, a square sub-matrix A ϵ ℝr×rof W with the same rank r as W is found. Then, the corresponding row restriction R of W is constructed. This leads to P = A-1ℝ and corresponding similarity matrix SW= (pTp)dmax, where dmaxis the maximum subspace dimension. Since most of the data matrices are low-rank in many subspace segmentation problems, this is computationally efficient compared to the other constructions of similarity matrices. It is also shown (with some limitations) that center-of-mass based sorting of data columns in SWcan be used to quickly assess clustering performance while algorithm development in both noisy or noise-free cases.

The motion segmentation problem is a special case of the general subspace segmentation problem that clusters data drawn from an unknown union of subspaces. This paper provides a nonlinear model for general subspace segmentation problem and presents an algorithm to compute the optimal solution for noiseless data. We also provide a combined algorithm that addresses issues with noise to some extent. Furthermore, a devised algorithm that specifically targets motion segmentation has been developed and applied to the Hopkins 155 Dataset. It generates the best segmentation rate to the date.

This letter presents a clustering algorithm for high dimensional data that comes from a union of lower dimensional subspaces of equal and known dimensions. The algorithm estimates a local subspace for each data point, and computes the distances between the local subspaces and the points to convert the problem to a one-dimensional data clustering problem. The algorithm is reliable in the presence of noise, and applied to the Hopkins 155 Dataset, it generates the best results to date for motion segmentation. The two motion, three motion, and overall segmentation rates for the video sequences are 99.43%, 98.69%, and 99.24%, respectively.

This paper presents a novel method for detecting multiple moving targets in real-time from infrared (IR) image sequences collected by an airborne IR camera. This novel method is based on dynamic Gabor filter and dynamic Gaussian detector. First, the ego-motion induced by the airborne platform is modeled by parametric affine transformation based on feature point matching, and the IR video is stabilized by eliminating the background motion. Then, a dynamic Gabor filter is employed to enhance the image changes for accurate detection and localization of moving targets. The orientation of Gabor filter is dynamically changed according to the orientation of optical flows. Next, the specular highlights generated by the dynamic Gabor filter are detected. The outliers and specular highlights are fused to identify the moving targets. The experimental results show that the proposed detection algorithm is effective and efficient. And the detection speed is approximate 2 frames per second.

Social robots need to have special human-robot interaction (HRI) systems to be accepted by people as natural partners. This paper first describes a multiple agent-based architecture designed to support HRI and then introduces an engagement mechanism based on attention distraction. The monitoring agent is a high-level agent that monitors and evaluates the surrounding environment and people. The interaction agent is another high-level agent that facilitates the seamless interaction between the robot and humans. The capability agent is a compound agent within the robot that is responsible for the robot's abilities and actions. The use of multiple HRI modalities in the engagement mechanism provides means for people to solicit the robot's attention while allowing the robot to ignore the distraction if it has recently become engaged with something else. The system has been successfully implemented and tested on a Pioneer 3-AT mobile robot

In this research, some human-robot interaction mechanisms that allow a human commander to control a mobile robot via a cellular phone have been developed and successfully tested. Two different control architectures have been implemented using different software technologies. The system includes two main modes, manual and autonomous controls. The manual control mode gives the user full control over the robot whereas the autonomous control mode allows the user to activate some built-in functions of the robot such as "navigate the meeting room". Two mobile robots, a Trilobot and a Pioneer, have been used as the development platforms in order to show the distinct features of mobile robot control with cellular phones.

In this research, a taxonomy is introduced to cover important considerations for human-robot interactions. As an application of passive human-robot interaction, two modalities for localizing humans based on sound source localization and infrared motion detection were developed and integrated with the face-tracker system of a humanoid ISAC (intelligent soft arm control), in order to direct ISACs attention and to prevent it from being quickly distracted. The sound source localization and passive infrared motion detection systems are used to provide the face-tracker system with candidate regions for finding a face. In order to avoid the situation where the robot appears to be "hyperactive" and cannot give sufficient attention to a newly discovered face, these sensing modules should not directly gain control of the tracking if the system has recently acquired a new face. Our goal is to allow a human to redirect the attention of the system but give the system a method to ignore the distraction if recently engaged.

The paper describes an adaptive genetic algorithm used to achieve multi-objectives such as minimizing the territory losses and maximizing enemy air losses by finding the optimum distribution of aircraft fighting in a war scenario simulated by the THUNDER software. The adaptive genetic algorithm changes the mutation and crossover adaptively to provide fast convergence to the optimum possible solutions. According to the population of the fitness values obtained for each generation, three distribution properties (the mean, the variance and the best fitness value) are determined and used as input to a fuzzy-logic system for modifying the mutation and crossover rates to obtain the individuals of the next generation. This enables fast and smooth convergence to the best possible solutions.

Robots are integrating more and more into our lives, However, to become an integral part of daily life, they should be socially accepted by humans. Evidently, the acceptance rate will increase as human-robot interaction gets closer to human-human interaction. The article describes the development of a Web based information filtering system, which enables a humanoid robot to initiate interaction with a human by generating human-like daily conversations.

The face tracker system of a humanoid, ISAC (Intelligent Soft Arm Control), is integrated with two modalities for localizing humans in order to direct ISAC's attention and to prevent ISAC from being quickly distracted. The sound source localization and passive infrared (PIR) motion detection systems are used to provide the face tracker system with candidate regions for finding a face. However, the sensing modules should not directly gain control of the tracking if the system has recently acquired a new face. Our goal is to allow a human to redirect the attention of the system, but give the system a method to ignore the distraction if recently engaged.