arXiv subjects

Farshad Khorrami

Publications and source records attributed to Farshad Khorrami.

At least 91 records · Page 5Linked to original sources

arXiv2023

R-LPIPS: An Adversarially Robust Perceptual Similarity Metric

Similarity metrics have played a significant role in computer vision to capture the underlying semantics of images. In recent years, advanced similarity metrics, such as the Learned Perceptual Image Patch Similarity (LPIPS), have emerged. These metrics leverage deep features extracted from trained neural networks and have demonstrated a remarkable ability to closely align with human perception when evaluating relative image similarity. However, it is now well-known that neural networks are susceptible to adversarial examples, i.e., small perturbations invisible to humans crafted to deliberately mislead the model. Consequently, the LPIPS metric is also sensitive to such adversarial examples. This susceptibility introduces significant security concerns, especially considering the widespread adoption of LPIPS in large-scale applications. In this paper, we propose the Robust Learned Perceptual Image Patch Similarity (R-LPIPS) metric, a new metric that leverages adversarially trained deep features. Through a comprehensive set of experiments, we demonstrate the superiority of R-LPIPS compared to the classical LPIPS metric. The code is available at https://github.com/SaraGhazanfari/R-LPIPS.

cs.CV↗

arXiv2023

Using Circulation to Mitigate Spurious Equilibria in Control Barrier Function -- Extended Version

Control Barrier Functions and Quadratic Programming are increasingly used for designing controllers that consider critical safety constraints. However, like Artificial Potential Fields, they can suffer from the stable spurious equilibrium point problem, which can result in the controller failing to reach the goal. To address this issue, we propose introducing circulation inequalities as a constraint. These inequalities force the system to explicitly circulate the obstacle region in configuration space, thus avoiding undesirable equilibria. We conduct a theoretical analysis of the proposed framework and demonstrate its efficacy through simulation studies. By mitigating spurious equilibria, our approach enhances the reliability of CBF-based controllers, making them more suitable for real-world applications.

eess.SY↗

arXiv2023

Differential Analysis of Triggers and Benign Features for Black-Box DNN Backdoor Detection

This paper proposes a data-efficient detection method for deep neural networks against backdoor attacks under a black-box scenario. The proposed approach is motivated by the intuition that features corresponding to triggers have a higher influence in determining the backdoored network output than any other benign features. To quantitatively measure the effects of triggers and benign features on determining the backdoored network output, we introduce five metrics. To calculate the five-metric values for a given input, we first generate several synthetic samples by injecting the input's partial contents into clean validation samples. Then, the five metrics are computed by using the output labels of the corresponding synthetic samples. One contribution of this work is the use of a tiny clean validation dataset. Having the computed five metrics, five novelty detectors are trained from the validation dataset. A meta novelty detector fuses the output of the five trained novelty detectors to generate a meta confidence score. During online testing, our method determines if online samples are poisoned or not via assessing their meta confidence scores output by the meta novelty detector. We show the efficacy of our methodology through a broad range of backdoor attacks, including ablation studies and comparison to existing approaches. Our methodology is promising since the proposed five metrics quantify the inherent differences between clean and poisoned samples. Additionally, our detection method can be incrementally improved by appending more metrics that may be proposed to address future advanced attacks.

cs.CR↗

arXiv2023

Towards Better Certified Segmentation via Diffusion Models

The robustness of image segmentation has been an important research topic in the past few years as segmentation models have reached production-level accuracy. However, like classification models, segmentation models can be vulnerable to adversarial perturbations, which hinders their use in critical-decision systems like healthcare or autonomous driving. Recently, randomized smoothing has been proposed to certify segmentation predictions by adding Gaussian noise to the input to obtain theoretical guarantees. However, this method exhibits a trade-off between the amount of added noise and the level of certification achieved. In this paper, we address the problem of certifying segmentation prediction using a combination of randomized smoothing and diffusion models. Our experiments show that combining randomized smoothing and diffusion models significantly improves certified robustness, with results indicating a mean improvement of 21 points in accuracy compared to previous state-of-the-art methods on Pascal-Context and Cityscapes public datasets. Our method is independent of the selected segmentation model and does not need any additional specialized training procedure.

cs.CV↗

arXiv2023

Data-Driven Deep Learning Based Feedback Linearization of Systems with Unknown Dynamics

A methodology is developed to learn a feedback linearization (i.e., nonlinear change of coordinates and input transformation) using a data-driven approach for a single input control-affine nonlinear system with unknown dynamics. We employ deep neural networks to learn the feedback law (input transformation) in conjunction with an extension of invertible neural networks to learn the nonlinear change of coordinates (state transformation). We also learn the matrices A and B of the transformed linear system and define loss terms to ensure controllability of the pair (A, B). The efficacy of our approach is demonstrated by simulations on several nonlinear systems. Furthermore, we show that state feedback controllers designed using the feedback linearized system yield expected closed-loop behavior when applied to the original nonlinear system, further demonstrating validity of the learned feedback linearization.

eess.SY↗

arXiv2023

State Constrained Stochastic Optimal Control for Continuous and Hybrid Dynamical Systems Using DFBSDE

We develop a computationally efficient learning-based forward-backward stochastic differential equations (FBSDE) controller for both continuous and hybrid dynamical (HD) systems subject to stochastic noise and state constraints. Solutions to stochastic optimal control (SOC) problems satisfy the Hamilton-Jacobi-Bellman (HJB) equation. Using current FBSDE-based solutions, the optimal control can be obtained from the HJB equations using deep neural networks (e.g., long short-term memory (LSTM) networks). To ensure the learned controller respects the constraint boundaries, we enforce the state constraints using a soft penalty function. In addition to previous works, we adapt the deep FBSDE (DFBSDE) control framework to handle HD systems consisting of continuous dynamics and a deterministic discrete state change. We demonstrate our proposed algorithm in simulation on a continuous nonlinear system (cart-pole) and a hybrid nonlinear system (five-link biped).

eess.SY↗

arXiv2023

Neural Lyapunov Control for Nonlinear Systems with Unstructured Uncertainties

Stabilizing controller design and region of attraction (RoA) estimation are essential in nonlinear control. Moreover, it is challenging to implement a control Lyapunov function (CLF) in practice when only partial knowledge of the system is available. We propose a learning framework that can synthesize state-feedback controllers and a CLF for control-affine nonlinear systems with unstructured uncertainties. Based on a regularity condition on these uncertainties, we model them as bounded disturbances and prove that a CLF for the nominal system (estimate of the true system) is an input-to-state stable control Lyapunov function (ISS-CLF) for the true system when the CLF's gradient is bounded. We integrate the robust Lyapunov analysis with the learning of both the control law and CLF. We demonstrate the effectiveness of our learning framework on several examples, such as an inverted pendulum system, a strict-feedback system, and a cart-pole system.

eess.SY↗

arXiv2023

Data-Efficient Control Barrier Function Refinement

Control barrier functions (CBFs) have been widely used for synthesizing controllers in safety-critical applications. When used as a safety filter, it provides a simple and computationally efficient way to obtain safe controls from a possibly unsafe performance controller. Despite its conceptual simplicity, constructing a valid CBF is well known to be challenging, especially for high-relative degree systems under nonconvex constraints. Recently, work has been done to learn a valid CBF from data based on a handcrafted CBF (HCBF). Even though the HCBF gives a good initialization point, it still requires a large amount of data to train the CBF network. In this work, we propose a new method to learn more efficiently from the collected data through a novel prioritized data sampling strategy. A priority score is computed from the loss value of each data point. Then, a probability distribution based on the priority score of the data points is used to sample data and update the learned CBF. Using our proposed approach, we can learn a valid CBF that recovers a larger portion of the true safe set using a smaller amount of data. The effectiveness of our method is demonstrated in simulation on a unicycle and a two-link arm.

eess.SY↗

arXiv2023

A Deep Neural Network Algorithm for Linear-Quadratic Portfolio Optimization with MGARCH and Small Transaction Costs

We analyze a fixed-point algorithm for reinforcement learning (RL) of optimal portfolio mean-variance preferences in the setting of multivariate generalized autoregressive conditional-heteroskedasticity (MGARCH) with a small penalty on trading. A numerical solution is obtained using a neural network (NN) architecture within a recursive RL loop. A fixed-point theorem proves that NN approximation error has a big-oh bound that we can reduce by increasing the number of NN parameters. The functional form of the trading penalty has a parameter $ε>0$ that controls the magnitude of transaction costs. When $ε$ is small, we can implement an NN algorithm based on the expansion of the solution in powers of $ε$. This expansion has a base term equal to a myopic solution with an explicit form, and a first-order correction term that we compute in the RL loop. Our expansion-based algorithm is stable, allows for fast computation, and outputs a solution that shows positive testing performance.

q-fin.CP↗

arXiv2022

Privacy-Preserving Collaborative Learning through Feature Extraction

We propose a framework in which multiple entities collaborate to build a machine learning model while preserving privacy of their data. The approach utilizes feature embeddings from shared/per-entity feature extractors transforming data into a feature space for cooperation between entities. We propose two specific methods and compare them with a baseline method. In Shared Feature Extractor (SFE) Learning, the entities use a shared feature extractor to compute feature embeddings of samples. In Locally Trained Feature Extractor (LTFE) Learning, each entity uses a separate feature extractor and models are trained using concatenated features from all entities. As a baseline, in Cooperatively Trained Feature Extractor (CTFE) Learning, the entities train models by sharing raw data. Secure multi-party algorithms are utilized to train models without revealing data or features in plain text. We investigate the trade-offs among SFE, LTFE, and CTFE in regard to performance, privacy leakage (using an off-the-shelf membership inference attack), and computational cost. LTFE provides the most privacy, followed by SFE, and then CTFE. Computational cost is lowest for SFE and the relative speed of CTFE and LTFE depends on network architecture. CTFE and LTFE provide the best accuracy. We use MNIST, a synthetic dataset, and a credit card fraud detection dataset for evaluations.

cs.LG↗

arXiv2022

Learning a Better Control Barrier Function

Control barrier functions (CBFs) are widely used in safety-critical controllers. However, constructing a valid CBF is challenging, especially under nonlinear or non-convex constraints and for high relative degree systems. Meanwhile, finding a conservative CBF that only recovers a portion of the true safe set is usually possible. In this work, starting from a "conservative" handcrafted CBF (HCBF), we develop a method to find a CBF that recovers a reasonably larger portion of the safe set. Since the learned CBF controller is not guaranteed to be safe during training iterations, we use a model predictive controller (MPC) to ensure safety during training. Using the collected trajectory data containing safe and unsafe interactions, we train a neural network to estimate the difference between the HCBF and a CBF that recovers a closer solution to the true safe set. With our proposed approach, we can generate safe controllers that are less conservative and computationally more efficient. We validate our approach on two systems: a second-order integrator and a ball-on-beam.

eess.SY↗

arXiv2022

Modular Multi-Copter Structure Control for Cooperative Aerial Cargo Transportation

The control problem of a multi-copter swarm, mechanically coupled through a modular lattice structure of connecting rods, is considered in this article. The system's structural elasticity is considered in deriving the system's dynamics. The devised controller is robust against the induced flexibilities, while an inherent adaptation scheme allows for the control of asymmetrical configurations and the transportation of unknown payloads. Certain optimization metrics are introduced for solving the individual agent thrust allocation problem while achieving maximum system flight time, resulting in a platform-independent control implementation. Experimental studies are offered to illustrate the efficiency of the suggested controller under typical flight conditions, increased rod elasticities and payload transportation.

cs.RO↗

arXiv2022

Matrix Pencil Based On-Line Computation of Controller Parameters in Dynamic High-Gain Scaling Controllers for Strict-Feedback Systems

We propose a new matrix pencil based approach for design of state-feedback and output-feedback stabilizing controllers for a general class of uncertain nonlinear strict-feedback-like systems. While the dynamic controller structure is based on the dual dynamic high-gain scaling based approach, we cast the design procedure within a general matrix pencil structure unlike prior results that utilized conservative algebraic bounds of terms arising in Lyapunov inequalities. The design freedoms in the dynamic controller are extracted in terms of generalized eigenvalues associated with matrix pencils formulated to capture the detailed structures (locations of uncertain terms in the system dynamics and their state dependences) of bounds in the Lyapunov analysis. The proposed approach enables efficient computation of non-conservative bounds with reduced algebraic complexity and enhances feasibility of application of the dual dynamic high-gain scaling based control designs. The proposed approach is developed under both the state-feedback and output-feedback cases and the efficacy of the approach is demonstrated through simulation studies on a numerical example.

math.OC↗

arXiv2022

Pop Quiz! Can a Large Language Model Help With Reverse Engineering?

Large language models (such as OpenAI's Codex) have demonstrated impressive zero-shot multi-task capabilities in the software domain, including code explanation. In this work, we examine if this ability can be used to help with reverse engineering. Specifically, we investigate prompting Codex to identify the purpose, capabilities, and important variable names or values from code, even when the code is produced through decompilation. Alongside an examination of the model's responses in answering open-ended questions, we devise a true/false quiz framework to characterize the performance of the language model. We present an extensive quantitative analysis of the measured performance of the language model on a set of program purpose identification and information extraction tasks: of the 136,260 questions we posed, it answered 72,754 correctly. A key takeaway is that while promising, LLMs are not yet ready for zero-shot reverse engineering.

cs.SE↗

arXiv2021

ESAFE: Enterprise Security and Forensics at Scale

Securing enterprise networks presents challenges in terms of both their size and distributed structure. Data required to detect and characterize malicious activities may be diffused and may be located across network and endpoint devices. Further, cyber-relevant data routinely exceeds total available storage, bandwidth, and analysis capability, often by several orders of magnitude. Real-time detection of threats within or across very large enterprise networks is not simply an issue of scale, but also a challenge due to the variable nature of malicious activities and their presentations. The system seeks to develop a hierarchy of cyber reasoning layers to detect malicious behavior, characterize novel attack vectors and present an analyst with a contextualized human-readable output from a series of machine learning models. We developed machine learning algorithms for scalable throughput and improved recall for our Multi-Resolution Joint Optimization for Enterprise Security and Forensics (ESAFE) solution. This Paper will provide an overview of ESAFE's Machine Learning Modules, Attack Ontologies, and Automated Smart Alert generation which provide multi-layer reasoning over cross-correlated sensors for analyst consumption.

cs.CR↗

arXiv2021

NNoculation: Catching BadNets in the Wild

This paper proposes a novel two-stage defense (NNoculation) against backdoored neural networks (BadNets) that, repairs a BadNet both pre-deployment and online in response to backdoored test inputs encountered in the field. In the pre-deployment stage, NNoculation retrains the BadNet with random perturbations of clean validation inputs to partially reduce the adversarial impact of a backdoor. Post-deployment, NNoculation detects and quarantines backdoored test inputs by recording disagreements between the original and pre-deployment patched networks. A CycleGAN is then trained to learn transformations between clean validation and quarantined inputs; i.e., it learns to add triggers to clean validation images. Backdoored validation images along with their correct labels are used to further retrain the pre-deployment patched network, yielding our final defense. Empirical evaluation on a comprehensive suite of backdoor attacks show that NNoculation outperforms all state-of-the-art defenses that make restrictive assumptions and only work on specific backdoor attacks, or fail on adaptive attacks. In contrast, NNoculation makes minimal assumptions and provides an effective defense, even under settings where existing defenses are ineffective due to attackers circumventing their restrictive assumptions.

cs.CR↗

arXiv2021

Prescribed-Time Regulation of Nonlinear Uncertain Systems with Unknown Input Gain and Appended Dynamics

The prescribed-time stabilization problem for a general class of nonlinear systems with unknown input gain and appended dynamics (with unmeasured state) is addressed. Unlike the asymptotic stabilization problem, the prescribed-time stabilization objective requires convergence of the state to the origin in a finite time that can be arbitrarily picked (i.e., prescribed) by the control system designer irrespective of the initial condition of the system. The class of systems considered is allowed to have general nonlinear uncertain terms throughout the system dynamics as well as uncertain appended dynamics (that effectively generate a time-varying non-vanishing disturbance signal input into the nominal system). The control design is based on a time scale transformation, dynamic high-gain scaling, and adaptation dynamics with temporal forcing terms.

math.OC↗

arXiv2021

Learning Locomotion Controllers for Walking Using Deep FBSDE

In this paper, we propose a deep forward-backward stochastic differential equation (FBSDE) based control algorithm for locomotion tasks. We also include state constraints in the FBSDE formulation to impose stable walking solutions or other constraints that one may want to consider (e.g., energy). Our approach utilizes a deep neural network (i.e., LSTM) to solve, in general, high-dimensional Hamilton-Jacobi-Bellman (HJB) equation resulting from the stated optimal control problem. As compared to traditional methods, our proposed method provides a higher computational efficiency in real-time; thus yielding higher frequency implementation of the closed-loop controllers. The efficacy of our approach is shown on a linear inverted pendulum model (LIPM) for walking. Even though we are deploying a simplified model of walking, the methodology is applicable to generalized and complex models for walking and other control/optimization tasks in robotic systems. Simulation studies have been provided to show the effectiveness of the proposed methodology.

cs.RO↗