Search

Deep neural networks have performed remarkably in many areas, including image-related classification tasks. However, various studies have shown that they are vulnerable to adversarial examples – images carefully crafted to fool well-trained deep neural networks by introducing imperceptible perturbations to the original images. To better understand the inherent characteristics of adversarial attacks, this paper studies the features of three common attack families: gradient-based, score-based, and decision-based. The primary objective is to recognize distinct types of adversarial examples, as identifying the type of information possessed by the attacker can aid in developing effective defense strategies. This paper demonstrates that adversarial images from different attack families can be successfully identified with a simple model. To further investigate the reason behind the observations, this paper conducts carefully designed experiments to study the distortion patterns of different attacks. Experimental results on CIFAR10 and Tiny ImageNet validated the differences in distortion patterns between various attack types for both $L_{2}$ and $L_{\infty } $ norm.

Mesenchymal stromal cells (MSCs) offer promising potential in biomedical research, clinical therapeutics, and immunomodulatory therapies due to their ease of isolation and multipotent, immunoprivileged, and immunosuppersive properties. Extensive efforts have focused on optimizing the cell isolation and culture methods to generate scalable, therapeutically-relevant MSCs for clinical applications. However, MSC-based therapies are often hindered by cell heterogeneity and inconsistency of therapeutic function caused, in part, by MSC senescence. As such, noninvasive and molecular-based MSC characterizations play an essential role in assuring the consistency of MSC functions. Here, we demonstrated that AI image translation algorithms can effectively predict immunofluorescence images of MSC senescence markers from phase contrast images. We showed that the expression level of senescence markers including senescence-associated beta-galactosidase (SABG), p16, p21, and p38 are accurately predicted by deep-learning models for Doxorubicin-induced MSC senescence, irradiation-induced MSC senescence, and replicative MSC senescence. Our AI model distinguished the non-senescent and senescent MSC populations and simultaneously captured the cell-to-cell variability within a population. Our microscopy-based phenotyping platform can be integrated with cell culture routines making it an easily accessible tool for MSC engineering and manufacturing.

Abstract Mesenchymal stromal cells (MSCs) are multipotent cells that have great potential for regenerative medicine, tissue repair, and immunotherapy. Unfortunately, the outcomes of MSC-based research and therapies can be highly inconsistent and difficult to reproduce, largely due to the inherently significant heterogeneity in MSCs, which has not been well investigated. To quantify cell heterogeneity, a standard approach is to measure marker expression on the protein level via immunochemistry assays. Performing such measurements non-invasively and at scale has remained challenging as conventional methods such as flow cytometry and immunofluorescence microscopy typically require cell fixation and laborious sample preparation. Here, we developed an artificial intelligence (AI)-based method that converts transmitted light microscopy images of MSCs into quantitative measurements of protein expression levels. By training a U-Net+ conditional generative adversarial network (cGAN) model that accurately (mean $$r_s$$ r s = 0.77) predicts expression of 8 MSC-specific markers, we showed that expression of surface markers provides a heterogeneity characterization that is complementary to conventional cell-level morphological analyses. Using this label-free imaging method, we also observed a multi-marker temporal-spatial fluctuation of protein distributions in live MSCs. These demonstrations suggest that our AI-based microscopy can be utilized to perform quantitative, non-invasive, single-cell, and multi-marker characterizations of heterogeneous live MSC culture. Our method provides a foundational step toward the instant integrative assessment of MSC properties, which is critical for high-throughput screening and quality control in cellular therapies.

Discrete combinatorial optimization problems in real world are typically defined via an ensemble of potentially high dimensional measurements pertaining to all subjects of a system under study. We point out that such a data ensemble in fact embeds with system's information content that is not directly used in defining the combinatorial optimization problems. Can machine learning algorithms extract such information content and make combinatorial optimizing tasks more efficient? Would such algorithmic computations bring new perspectives into this classic topic of Applied Mathematics and Theoretical Computer Science? We show that answers to both questions are positive. One key reason is due to permutation invariance. That is, the data ensemble of subjects' measurement vectors is permutation invariant when it is represented through a subject-vs-measurement matrix. An unsupervised machine learning algorithm, called Data Mechanics (DM), is applied to find optimal permutations on row and column axes such that the permuted matrix reveals coupled deterministic and stochastic structures as the system's information content. The deterministic structures are shown to facilitate geometry-based divide-and-conquer scheme that helps optimizing task, while stochastic structures are used to generate an ensemble of mimicries retaining the deterministic structures, and then reveal the robustness pertaining to the original version of optimal solution. Two simulated systems, Assignment problem and Traveling Salesman problem, are considered. Beyond demonstrating computational advantages and intrinsic robustness in the two systems, we propose brand new robust optimal solutions. We believe such robust versions of optimal solutions are potentially more realistic and practical in real world settings.