Shadows can be Dangerous: Stealthy and Effective Physical-world Adversarial Attack by Natural Phenomenon

Estimating the risk level of adversarial examples is essential for safely deploying machine learning models in the real world. One popular approach for physical-world attacks is to adopt the "sticker-pasting" strategy, which however suffers from some limitations, including difficulties in access to the target or printing by valid colors. A new type of non-invasive attacks emerged recently, which attempt to cast perturbation onto the target by optics based tools, such as laser beam and projector. However, the added optical patterns are artificial but not natural. Thus, they are still conspicuous and attention-grabbed, and can be easily noticed by humans. In this paper, we study a new type of optical adversarial examples, in which the perturbations are generated by a very common natural phenomenon, shadow, to achieve naturalistic and stealthy physical-world adversarial attack under the black-box setting. We extensively evaluate the effectiveness of this new attack on both simulated and real-world environments. Experimental results on traffic sign recognition demonstrate that our algorithm can generate adversarial examples effectively, reaching 98.23% and 90.47% success rates on LISA and GTSRB test sets respectively, while continuously misleading a moving camera over 95% of the time in real-world scenarios. We also offer discussions about the limitations and the defense mechanism of this attack.

On the Convexity of Discrete Time Covariance Steering in Stochastic Linear Systems with Wasserstein Terminal Cost

In this work, we analyze the properties of the solution to the covariance steering problem for discrete time Gaussian linear systems with a squared Wasserstein distance terminal cost. In our previous work, we have shown that by utilizing the state feedback control policy parametrization, this stochastic optimal control problem can be associated with a difference of convex functions program. Here, we revisit the same covariance control problem but this time we focus on the analysis of the problem. Specifically, we establish the existence of solutions to the optimization problem and derive the first and second order conditions for optimality. We provide analytic expressions for the gradient and the Hessian of the performance index by utilizing specialized tools from matrix calculus. Subsequently, we prove that the optimization problem always admits a global minimizer, and finally, we provide a sufficient condition for the performance index to be a strictly convex function (under the latter condition, the problem admits a unique global minimizer). In particular, we show that when the terminal state covariance is upper bounded, with respect to the L\"{o}wner partial order, by the covariance matrix of the desired terminal normal distribution, then our problem admits a unique global minimizing state feedback gain. The results of this paper set the stage for the development of specialized control design tools that exploit the structure of the solution to the covariance steering problem with a squared Wasserstein distance terminal cost.

Time-Variant Variational Transfer for Value Functions

In most of the transfer learning approaches to reinforcement learning (RL) the distribution over the tasks is assumed to be stationary. Therefore, the target and source tasks are i.i.d. samples of the same distribution. In the context of this work, we consider the problem of transferring value functions through a variational method when the distribution that generates the tasks is time-variant, proposing a solution that leverages this temporal structure inherent in the task generating process. Furthermore, by means of a finite-sample analysis, the previously mentioned solution is theoretically compared to its time-invariant version. Finally, we will provide an experimental evaluation of the proposed technique with three distinct temporal dynamics in three different RL environments.

Domain Generalization using Pretrained Models without Fine-tuning

Fine-tuning pretrained models is a common practice in domain generalization (DG) tasks. However, fine-tuning is usually computationally expensive due to the ever-growing size of pretrained models. More importantly, it may cause over-fitting on source domain and compromise their generalization ability as shown in recent works. Generally, pretrained models possess some level of generalization ability and can achieve decent performance regarding specific domains and samples. However, the generalization performance of pretrained models could vary significantly over different test domains even samples, which raises challenges for us to best leverage pretrained models in DG tasks. In this paper, we propose a novel domain generalization paradigm to better leverage various pretrained models, named specialized ensemble learning for domain generalization (SEDGE). It first trains a linear label space adapter upon fixed pretrained models, which transforms the outputs of the pretrained model to the label space of the target domain. Then, an ensemble network aware of model specialty is proposed to dynamically dispatch proper pretrained models to predict each test sample. Experimental studies on several benchmarks show that SEDGE achieves significant performance improvements comparing to strong baselines including state-of-the-art method in DG tasks and reduces the trainable parameters by ~99% and the training time by ~99.5%.

Sky Computing: Accelerating Geo-distributed Computing in Federated Learning

Federated learning is proposed by Google to safeguard data privacy through training models locally on users' devices. However, with deep learning models growing in size to achieve better results, it becomes increasingly difficult to accommodate the whole model on one single device. Thus, model parallelism is then used to divide the model weights among several devices. With this logic, the approach currently used evenly allocates weights among devices. However, in reality, a computation bottleneck may occur resulting from variant computing power of different users' devices. To address this problem, load balancing is needed to allocate the model weights based on the computational capability of the device. In this paper, we proposed Sky Computing, a load-balanced model parallelism framework to adaptively allocate the weights to devices. Sky Computing outperforms the baseline method by 55% in training time when training 160-layer BERT with 64 nodes. The source code can be found at https://github.com/hpcaitech/SkyComputing.

Finite-time disturbance reconstruction and robust fractional-order controller design for hybrid port-Hamiltonian dynamics of biped robots

In this paper, disturbance reconstruction and robust trajectory tracking control of biped robots with hybrid dynamics in the port-Hamiltonian form is investigated. A new type of Hamiltonian function is introduced, which ensures the finite-time stability of the closed-loop system. The proposed control system consists of two loops: an inner and an outer loop. A fractional proportional-integral-derivative filter is used to achieve finite-time convergence for position tracking errors at the outer loop. A fractional-order sliding mode controller acts as a centralized controller at the inner-loop, ensuring the finite-time stability of the velocity tracking error. In this loop, the undesired effects of unknown external disturbance and parameter uncertainties are compensated using estimators. Two disturbance estimators are envisioned. The former is designed using fractional calculus. The latter is an adaptive estimator, and it is constructed using the general dynamic of biped robots. Stability analysis shows that the closed-loop system is finite-time stable in both contact-less and impact phases. Simulation studies on two types of biped robots (i.e., two-link walker and RABBIT biped robot) demonstrate the proposed controller's tracking performance and disturbance rejection capability.

MediaPipe Hands: On-device Real-time Hand Tracking

We present a real-time on-device hand tracking pipeline that predicts hand skeleton from single RGB camera for AR/VR applications. The pipeline consists of two models: 1) a palm detector, 2) a hand landmark model. It's implemented via MediaPipe, a framework for building cross-platform ML solutions. The proposed model and pipeline architecture demonstrates real-time inference speed on mobile GPUs and high prediction quality. MediaPipe Hands is open sourced at https://mediapipe.dev.

Typography-MNIST (TMNIST): an MNIST-Style Image Dataset to Categorize Glyphs and Font-Styles

We present Typography-MNIST (TMNIST), a dataset comprising of 565,292 MNIST-style grayscale images representing 1,812 unique glyphs in varied styles of 1,355 Google-fonts. The glyph-list contains common characters from over 150 of the modern and historical language scripts with symbol sets, and each font-style represents varying subsets of the total unique glyphs. The dataset has been developed as part of the CognitiveType project which aims to develop eye-tracking tools for real-time mapping of type to cognition and to create computational tools that allow for the easy design of typefaces with cognitive properties such as readability. The dataset and scripts to generate MNIST-style images for glyphs in different font styles are freely available at https://github.com/aiskunks/CognitiveType.

Deciding cuspidality of manipulators through computer algebra and algorithms in real algebraic geometry

Cuspidal robots are robots with at least two inverse kinematic solutions that can be connected by a singularity-free path. Deciding the cuspidality of a generic 3R robots has been studied in the past, but extending the study to six-degree-of-freedom robots can be a challenging problem. Many robots can be modeled as a polynomial map together with a real algebraic set so that the notion of cuspidality can be extended to these data.In this paper we design an algorithm that, on input a polynomial map in $n$ indeterminates, and $s$ polynomials in the same indeterminates describing a real algebraic set of dimension $d$, decides the cuspidality of the restriction of the map to the real algebraic set under consideration. Moreover, if $D$ and $\tau$ are respectively the maximum degree and a bound on the bit size of the coefficients of the input polynomials, this algorithm runs in time log-linear in $\tau$ and polynomial in $((s+d)nD)^{O(n^2)}$.It relies on many high-level algorithms in computer algebra which use advanced methods on real algebraic sets and critical loci of polynomial maps. As far as we know, this is the first algorithm that tackles the cuspidality problem from a general point of view.

Dynamic Window-level Granger Causality of Multi-channel Time Series

Granger causality method analyzes the time series causalities without building a complex causality graph. However, the traditional Granger causality method assumes that the causalities lie between time series channels and remain constant, which cannot model the real-world time series data with dynamic causalities along the time series channels. In this paper, we present the dynamic window-level Granger causality method (DWGC) for multi-channel time series data. We build the causality model on the window-level by doing the F-test with the forecasting errors on the sliding windows. We propose the causality indexing trick in our DWGC method to reweight the original time series data. Essentially, the causality indexing is to decrease the auto-correlation and increase the cross-correlation causal effects, which improves the DWGC method. Theoretical analysis and experimental results on two synthetic and one real-world datasets show that the improved DWGC method with causality indexing better detects the window-level causalities.