no code implementations • 10 Mar 2024 • Ziye Ma, Ying Chen, Javad Lavaei, Somayeh Sojoudi
Matrix sensing problems exhibit pervasive non-convexity, plaguing optimization with a proliferation of suboptimal spurious solutions.
no code implementations • 15 Feb 2024 • Jingqi Li, Anand Siththaranjan, Somayeh Sojoudi, Claire Tomlin, Andrea Bajcsy
Autonomous agents should be able to coordinate with other agents without knowing their intents ahead of time.
1 code implementation • 3 Feb 2024 • Yatong Bai, Mo Zhou, Vishal M. Patel, Somayeh Sojoudi
Adversarial robustness often comes at the cost of degraded accuracy, impeding the real-life application of robust classification models.
no code implementations • 9 Jan 2024 • Yatong Bai, Utsav Garg, Apaar Shanker, Haoming Zhang, Samyak Parajuli, Erhan Bas, Isidora Filipovic, Amelia N. Chu, Eugenia D Fomitcheva, Elliot Branson, Aerin Kim, Somayeh Sojoudi, Kyunghyun Cho
Vision and vision-language applications of neural networks, such as image classification and captioning, rely on large-scale annotated datasets that require non-trivial data-collecting processes.
no code implementations • 26 Nov 2023 • Yatong Bai, Brendon G. Anderson, Somayeh Sojoudi
However, standard learning models often suffer from an accuracy-robustness trade-off, which is a limitation that must be overcome in the control of safety-critical systems that require both high performance and rigorous robustness guarantees.
no code implementations • 7 Oct 2023 • Brendon G. Anderson, Samuel Pfrommer, Somayeh Sojoudi
The reliable deployment of neural networks in control systems requires rigorous robustness guarantees.
no code implementations • 4 Oct 2023 • Tanmay Gautam, Reid Pryzant, ZiYi Yang, Chenguang Zhu, Somayeh Sojoudi
SCQ works like a differentiable convex optimization (DCO) layer: in the forward pass, we solve for the optimal convex combination of codebook vectors that quantize the inputs.
1 code implementation • 25 Sep 2023 • Samuel Pfrommer, Brendon G. Anderson, Somayeh Sojoudi
Randomized smoothing is the current state-of-the-art method for producing provably robust classifiers.
1 code implementation • 19 Sep 2023 • Yatong Bai, Trung Dang, Dung Tran, Kazuhito Koishida, Somayeh Sojoudi
Diffusion models power a vast majority of text-to-audio (TTA) generation methods.
Ranked #10 on Audio Generation on AudioCaps
no code implementations • 29 Jul 2023 • Samuel Pfrommer, Yatong Bai, Hyunin Lee, Somayeh Sojoudi
Imitation learning suffers from causal confusion.
no code implementations • 4 Apr 2023 • Jingqi Li, Chih-Yuan Chiu, Lasse Peters, Fernando Palafox, Mustafa Karabag, Javier Alonso-Mora, Somayeh Sojoudi, Claire Tomlin, David Fridovich-Keil
To accommodate this, we decompose the approximated game into a set of smaller games with few constraints for each sampled scenario, and propose a decentralized, consensus-based ADMM algorithm to efficiently compute a generalized Nash equilibrium (GNE) of the approximated game.
no code implementations • 29 Mar 2023 • Tanmay Gautam, Samuel Pfrommer, Somayeh Sojoudi
Conventional optimization methods in machine learning and controls rely heavily on first-order update rules.
no code implementations • 15 Feb 2023 • Ziye Ma, Igor Molybog, Javad Lavaei, Somayeh Sojoudi
This paper studies the role of over-parametrization in solving non-convex optimization problems.
1 code implementation • 29 Jan 2023 • Yatong Bai, Brendon G. Anderson, Aerin Kim, Somayeh Sojoudi
While prior research has proposed a plethora of methods that build neural classifiers robust against adversarial robustness, practitioners are still reluctant to adopt them due to their unacceptably severe clean accuracy penalties.
Ranked #1 on Adversarial Robustness on CIFAR-100 (using extra training data)
1 code implementation • 28 Oct 2022 • Fengjun Yang, Fernando Gama, Somayeh Sojoudi, Nikolai Matni
Designing distributed optimal controllers subject to communication constraints is a difficult problem unless structural assumptions are imposed on the underlying dynamics and information exchange structure, e. g., sparsity, delay, or spatial invariance.
no code implementations • 21 Sep 2022 • Samuel Pfrommer, Somayeh Sojoudi
Under mild conditions, we show that the disturbance-aware policy converges to the blind online policy if the number of disturbances grows sublinearly in the time horizon.
no code implementations • 15 Aug 2022 • Baturalp Yalcin, Ziye Ma, Javad Lavaei, Somayeh Sojoudi
In this paper, we shed light on some major differences between these two methods.
no code implementations • 15 Aug 2022 • Brendon G. Anderson, Tanmay Gautam, Somayeh Sojoudi
In this discussion paper, we survey recent research surrounding robustness of machine learning models.
no code implementations • 18 Mar 2022 • Jingqi Li, Donggun Lee, Somayeh Sojoudi, Claire J. Tomlin
We address this problem by designing a new value function with a contracting Bellman backup, where the super-zero level set, i. e., the set of states where the value function is evaluated to be non-negative, recovers the reach-avoid set.
1 code implementation • 8 Mar 2022 • Ziye Ma, Somayeh Sojoudi
We prove that as long as the RIP constant of the noiseless objective is less than $1/3$, any spurious local solution of the noisy optimization problem must be close to the ground truth solution.
no code implementations • 6 Jan 2022 • Yatong Bai, Tanmay Gautam, Somayeh Sojoudi
We apply the robust convex optimization theory to convex training and develop convex formulations that train ANNs robust to adversarial inputs.
no code implementations • 27 Dec 2021 • Samuel Pfrommer, Tanmay Gautam, Alec Zhou, Somayeh Sojoudi
Real-world reinforcement learning (RL) problems often demand that agents behave safely by obeying a set of designed constraints.
no code implementations • 19 Oct 2021 • Baturalp Yalcin, Haixiang Zhang, Javad Lavaei, Somayeh Sojoudi
It is well-known that the Burer-Monteiro (B-M) factorization approach can efficiently solve low-rank matrix optimization problems under the RIP condition.
no code implementations • 31 May 2021 • Fernando Gama, Brendon G. Anderson, Somayeh Sojoudi
We show that, by replacing nonlinear activation functions by NVGFs, frequency creation mechanisms can be designed or learned.
no code implementations • 25 May 2021 • Yatong Bai, Tanmay Gautam, Yu Gai, Somayeh Sojoudi
Recent work has shown that the training of a one-hidden-layer, scalar-output fully-connected ReLU neural network can be reformulated as a finite-dimensional convex program.
no code implementations • 18 May 2021 • Ziye Ma, Yingjie Bi, Javad Lavaei, Somayeh Sojoudi
By analyzing the landscape of the non-convex problem, we first propose a global guarantee on the maximum distance between an arbitrary local minimizer and the ground truth under the assumption that the RIP constant is smaller than $1/2$.
no code implementations • 15 Mar 2021 • Fernando Gama, Somayeh Sojoudi
Controlling network systems has become a problem of paramount importance.
no code implementations • 22 Jan 2021 • Brendon G. Anderson, Ziye Ma, Jingqi Li, Somayeh Sojoudi
We extend the analysis to the SDP, where the feasible set geometry is exploited to design a branching scheme that minimizes the worst-case SDP relaxation error.
no code implementations • 7 Dec 2020 • Mohammad Mahdi Khalili, Xueru Zhang, Mahed Abroshan, Somayeh Sojoudi
In this work, we study the possibility of using a differentially private exponential mechanism as a post-processing step to improve both fairness and privacy of supervised learning models.
no code implementations • 10 Nov 2020 • Fernando Gama, Somayeh Sojoudi
When considering a network system, this renders the optimal controller a centralized one.
no code implementations • 16 Oct 2020 • Ziye Ma, Somayeh Sojoudi
We analyze the performance of this sequential SDP method both theoretically and empirically, and show that it bridges the gap as the number of cuts increases.
no code implementations • 15 Oct 2020 • Brendon G. Anderson, Somayeh Sojoudi
Methods to certify the robustness of neural networks in the presence of input uncertainty are vital in safety-critical settings.
no code implementations • 2 Oct 2020 • Brendon G. Anderson, Somayeh Sojoudi
Methods to certify the robustness of neural networks in the presence of input uncertainty are vital in safety-critical settings.
1 code implementation • NeurIPS 2020 • Fangda Gu, Heng Chang, Wenwu Zhu, Somayeh Sojoudi, Laurent El Ghaoui
Graph Neural Networks (GNNs) are widely used deep learning models that learn meaningful representations from graph-structured data.
no code implementations • 1 Apr 2020 • Brendon G. Anderson, Ziye Ma, Jingqi Li, Somayeh Sojoudi
In this paper, we consider the problem of certifying the robustness of neural networks to perturbed and adversarial input data.
no code implementations • 16 Mar 2020 • Heng Chang, Yu Rong, Tingyang Xu, Wenbing Huang, Somayeh Sojoudi, Junzhou Huang, Wenwu Zhu
In this paper, we first introduce the attention mechanism in the spectral domain of graphs and present Spectral Graph Attention Network (SpGAT) that learns representations for different frequency components regarding weighted filters and graph wavelets bases.
no code implementations • 25 Sep 2019 • Ming Jin, Heng Chang, Wenwu Zhu, Somayeh Sojoudi
By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability.
no code implementations • 25 Sep 2019 • Heng Chang, Yu Rong, Somayeh Sojoudi, Junzhou Huang, Wenwu Zhu
Many variants of Graph Convolutional Networks (GCNs) for representation learning have been proposed recently and have achieved fruitful results in various domains.
no code implementations • 21 Sep 2019 • Salar Fattahi, Nikolai Matni, Somayeh Sojoudi
In this work, we propose a robust approach to design distributed controllers for unknown-but-sparse linear and time-invariant systems.
no code implementations • 9 Jul 2019 • Brendon G. Anderson, Somayeh Sojoudi
In this paper, we consider the problem of unsupervised video object segmentation via background subtraction.
1 code implementation • 24 May 2019 • Ming Jin, Heng Chang, Wenwu Zhu, Somayeh Sojoudi
By challenging the classical graph Laplacian, we propose a new convolution operator that is provably robust in the spectral domain and is incorporated in the GCN architecture to improve expressivity and interpretability.
no code implementations • 20 Apr 2019 • Salar Fattahi, Nikolai Matni, Somayeh Sojoudi
In particular, we show that the proposed estimator can correctly identify the sparsity pattern of the system matrices with high probability, provided that the length of the sample trajectory exceeds a threshold.
no code implementations • 7 Jan 2019 • Richard Y. Zhang, Somayeh Sojoudi, Javad Lavaei
Using the technique, we prove that in the case of a rank-1 ground truth, an RIP constant of $\delta<1/2$ is both necessary and sufficient for exact recovery from any arbitrary initial point (such as a random point).
no code implementations • 30 Dec 2018 • Salar Fattahi, Somayeh Sojoudi
In particular, it is shown that a constant fraction of the measurements could be grossly corrupted and yet they would not create any spurious local solution.
no code implementations • 16 Sep 2018 • Nan Tian, Jinfa Chen, Mas Ma, Robert Zhang, Bill Huang, Ken Goldberg, Somayeh Sojoudi
We use the system to enable robust teleoperation of a dynamic self-balancing robot from the cloud.
no code implementations • NeurIPS 2018 • Richard Y. Zhang, Cédric Josz, Somayeh Sojoudi, Javad Lavaei
When the linear measurements of an instance of low-rank matrix recovery satisfy a restricted isometry property (RIP)---i. e. they are approximately norm-preserving---the problem is known to contain no spurious local minima, so exact recovery is guaranteed.
no code implementations • 21 Mar 2018 • Salar Fattahi, Somayeh Sojoudi
A by-product of this result is that the number of sample trajectories required for sparse system identification is significantly smaller than the dimension of the system.
no code implementations • ICML 2018 • Richard Y. Zhang, Salar Fattahi, Somayeh Sojoudi
The sparse inverse covariance estimation problem is commonly solved using an $\ell_{1}$-regularized Gaussian maximum likelihood estimator known as "graphical lasso", but its computational cost becomes prohibitive for large data sets.
no code implementations • 24 Nov 2017 • Salar Fattahi, Richard Y. Zhang, Somayeh Sojoudi
We have also derived a closed-form solution that is optimal when the thresholded sample covariance matrix has an acyclic structure.
no code implementations • 30 Aug 2017 • Salar Fattahi, Somayeh Sojoudi
The objective of this paper is to compare the computationally-heavy GL technique with a numerically-cheap heuristic method that is based on simply thresholding the sample covariance matrix.