Search

Adaptive Deep Brain Stimulation (aDBS) has been proposed in literature to avoid the negative consequences associated with the continuous stimulation delivered through traditional deep brain stimulation. This work seeks to determine a group of neural biomarkers that a classification algorithm could use on an aDBS device using rodent animal models. The neural activities were acquired from the primary motor cortex of four Parkinsonian model rats and four healthy rats from a control group. To overcome the variability introduced from the small rat sample size, this work proposes a novel method for combining and running Genetic Feature Selection and Forward Stepwise Feature Selection in an environment where classification accuracy varies greatly based on how the folds are organized before cross-validation. Three separate classification algorithms, Logistic Regression, k-Nearest Neighbor, and Random Forest are used to verify the proposed method. For Logistic Regression, the set of Alpha Power (7-12 Hz), High Beta Power (20-30 Hz), and 55-95 Hz Gamma Power shows the best performance in classification. For k-Nearest Neighbor, the characterizing features are Low Beta Power (12-20 Hz), High Beta Power, All Beta Power (12-30 Hz), 55-95 Hz Gamma Power, and 95-105 Hz Gamma Power. For Random Forest, they are High Beta Power, All Beta Power, 55-95 Hz Gamma Power, 95-105 Hz Gamma Power, and 300-350 Hz High-Frequency Oscillations Power. With the selected feature set, experimental results show an increasing classification accuracy from 59.08% to 77.69% for Logistic Regression, from 49.53% to 73.44% for k-Nearest Neighbor, and from 54.10% to 71.15% for Random Forest. Clinical Relevance- This experiment provides a method for determining the most effective biomarkers from a larger set for classifying Parkinsonian behavior for an aDBS device.

This paper reports the design and experimental characterization of a cryogenic compact low-power 60GHz amplifier for control of electron/hole spin qubits, as elementary building block for monolithic Si quantum processors. Tested at 2 K, the amplifier exhibits $S$ 21 of 15 dB at 59 GHz, BW3dB of 52.5-67.5 GHz, and power consumption of 2.16 mW. Owing to the topology with inductorless active network, the amplifier has a compact core area of O.18×O.19 mm2.

Single image generative models perform synthesis and manipulation tasks by capturing the distribution of patches within a single image. The classical (pre Deep Learning) prevailing approaches for these tasks are based on an optimization process that maximizes patch similarity between the input and generated output. Recently, however, Single Image GANs were introduced both as a superior solution for such manipulation tasks, but also for remarkable novel generative tasks. Despite their impressiveness, single image GANs require long training time (usually hours) for each image and each task. They often suffer from artifacts and are prone to optimization issues such as mode collapse. In this paper, we show that all of these tasks can be performed without any training, within several seconds, in a unified, surprisingly simple framework. We revisit and cast the "good-old" patch-based methods into a novel optimization-free framework. We start with an initial coarse guess, and then simply refine the details coarse-to-fine using patch-nearest-neighbor search. This allows generating random novel images better and much faster than GANs. We further demonstrate a wide range of applications, such as image editing and reshuffling, retargeting to different sizes, structural analogies, image collage and a newly introduced task of conditional inpainting. Not only is our method faster ($\times 10^3$-$\times 10^4$ than a GAN), it produces superior results (confirmed by quantitative and qualitative evaluation), less artifacts and more realistic global structure than any of the previous approaches (whether GAN-based or classical patch-based)., 11 pages, 10 figures, added references and acknowledgments

Anomaly detection is a well-established research area that seeks to identify samples outside of a predetermined distribution. An anomaly detection pipeline is comprised of two main stages: (1) feature extraction and (2) normality score assignment. Recent papers used pre-trained networks for feature extraction achieving state-of-the-art results. However, the use of pre-trained networks does not fully-utilize the normal samples that are available at train time. This paper suggests taking advantage of this information by using teacher-student training. In our setting, a pretrained teacher network is used to train a student network on the normal training samples. Since the student network is trained only on normal samples, it is expected to deviate from the teacher network in abnormal cases. This difference can serve as a complementary representation to the pre-trained feature vector. Our method -- Transformaly -- exploits a pre-trained Vision Transformer (ViT) to extract both feature vectors: the pre-trained (agnostic) features and the teacher-student (fine-tuned) features. We report state-of-the-art AUROC results in both the common unimodal setting, where one class is considered normal and the rest are considered abnormal, and the multimodal setting, where all classes but one are considered normal, and just one class is considered abnormal. The code is available at https://github.com/MatanCohen1/Transformaly., Comment: CVPR Workshop, 2022

This paper reports the experimental characterization at the cryogenic temperature of a compact mm-wave broadband single-pole single-throw (SPST) switch in 22nm FDSOI CMOS technology. The switch consists of two n-MOSFETs with a special device option to reduce the substrate parasitic effects, and a third n-MOSFET to improve isolation. Unlike prior wideband mm-wave switches, it does not require any large passive components, allowing a very compact design, low loss and high isolation performance. The cryogenic measurements at 2 K show an insertion loss lower than 2.3 dB, an isolation better than 25.3 dB, and the return loss better than -11.5 dB, over the entire frequency range from DC to 70 GHz.

We present a new method for real-time non-rigid dense correspondence between point clouds based on structured shape construction. Our method, termed Deep Point Correspondence (DPC), requires a fraction of the training data compared to previous techniques and presents better generalization capabilities. Until now, two main approaches have been suggested for the dense correspondence problem. The first is a spectral-based approach that obtains great results on synthetic datasets but requires mesh connectivity of the shapes and long inference processing time while being unstable in real-world scenarios. The second is a spatial approach that uses an encoder-decoder framework to regress an ordered point cloud for the matching alignment from an irregular input. Unfortunately, the decoder brings considerable disadvantages, as it requires a large amount of training data and struggles to generalize well in cross-dataset evaluations. DPC's novelty lies in its lack of a decoder component. Instead, we use latent similarity and the input coordinates themselves to construct the point cloud and determine correspondence, replacing the coordinate regression done by the decoder. Extensive experiments show that our construction scheme leads to a performance boost in comparison to recent state-of-the-art correspondence methods. Our code is publicly available at https://github.com/dvirginz/DPC., 3DV 2021

Deep neural networks are prone to adversarial examples that maliciously alter the network's outcome. Due to the increasing popularity of 3D sensors in safety-critical systems and the vast deployment of deep learning models for 3D point sets, there is a growing interest in adversarial attacks and defenses for such models. So far, the research has focused on the semantic level, namely, deep point cloud classifiers. However, point clouds are also widely used in a geometric-related form that includes encoding and reconstructing the geometry. In this work, we are the first to consider the problem of adversarial examples at a geometric level. In this setting, the question is how to craft a small change to a clean source point cloud that leads, after passing through an autoencoder model, to the reconstruction of a different target shape. Our attack is in sharp contrast to existing semantic attacks on 3D point clouds. While such works aim to modify the predicted label by a classifier, we alter the entire reconstructed geometry. Additionally, we demonstrate the robustness of our attack in the case of defense, where we show that remnant characteristics of the target shape are still present at the output after applying the defense to the adversarial input. Our code is publicly available at https://github.com/itailang/geometric_adv., 3DV 2021

Density is a crucial indicator for the level of services on highways. Due to the limitations of monitoring devices, speed, occupancy, and volume to capacity are used as surrogates to help authorities determine road congestion. Attempts have been made to utilize computer vision-based methods to retrieve accurate road density estimation. However, two significant problems emerge for CV-based methods: 1) Features of vehicles are not clear enough to be recognized in low-resolution or dense scenes. 2) The lack of knowledge of road length hinders the density calculation. This paper intends to address these limitations by developing a congestion detection framework, including the Vehicle-Counter Convolutional Neural Network (VC-CNN) and the Self-Learning Critical Density Component (SL-CDC). The VC-CNN produces estimations of vehicle count by creating accurate density maps. The SL-CDC divides the road section into two homogeneous cells and distinguishes the dynamic characteristics of traffic flow, including shockwaves, to derive the critical density of the road section and identify forming congestion. This framework has been validated with a wide range of experiments and applied to field tests on the G2 Expressway in China with considerable accuracy and reliability, satisfying the demand of the transportation authorities to monitor and identify the forming congestion.

Autonomous driving is disrupting the automotive industry as we know it today. For this, fail-operational behavior is essential in the sense, plan, and act stages of the automation chain in order to handle safety-critical situations on its own, which currently is not reached with state-of-the-art approaches.The European ECSEL research project PRYSTINE realizes Fail-operational Urban Surround perceptION (FUSION) based on robust Radar and LiDAR sensor fusion and control functions in order to enable safe automated driving in urban and rural environments. This paper showcases some of the key exploitable results (e.g., novel Radar sensors, innovative embedded control and E/E architectures, pioneering sensor fusion approaches, AI-controlled vehicle demonstrators) achieved until its final year 3.

While automotive research plays a significant role nowadays, most of the experimental activity demands costly platforms and involves safety issues. Plenty of driving simulators were proposed to reduce the costs and guarantee safety. However, they still cannot reflect the physical world, resulting in subjective assessments in any aspect of the study. This paper introduces an affordable remote driving testbed based on small-scale car-like mobile platforms and a physical road. The driver in the remote driving station observes a real-time video taken from a front-facing camera installed in the car. For a realistic driving experience, we have developed a torque feedback mechanism based on the small-scale car motion to mimic the influence of the physical linkage between the front wheels and the steering wheel of a standard car. This mechanism demands knowledge of the car’s side-slip angle that is not directly measured. Here, we introduce a supervised learning-based combined regression model (RidgeCV and Bootstrap aggregating decision tree) that estimates the side-slip angle for highly non-linear behavior.

The article addresses the tracking control problem of differentially driven nonholonomic Wheeled Mobile Robots (WMRs) whose nonlinear kinematic-dynamic equations of motion subject to bounded inputs. To overcome the difficulties arising from the constraints in the system and to avoid singular states, the potential for their appearance is common in this system, we implement sigmoid functions in the feedback loop. Also, since the speed variables (linear and angular) that enter the kinematic subsystem are not independent variables and are the result of the action of bounded forces/torques on the dynamic system, we present a special way of reconstructing them. Using the flatness property of the system, the article presents a control algorithm for tracking a time-parameterized path (a geometric path with an associated timing-law). The proposed control scheme allows us to present conditions on the required trajectory that, if they are met, the input signals satisfy the given restrictions and ensures convergence to the reference trajectory.

We present a divide and conquer design approach for handling control problems with dynamically changing objectives. The approach is based on associating each objective with a virtual input and then considering a game between a set of (virtual) players, each trying to achieve an objective by controlling the associated virtual input. By guarantying a Nash equilibrium for this game, we synthesize a combined controller that establishes a stable balance between the objectives. To demonstrate the use of the approach, we describe how a controller for a non-linear system (quadrotor) can be designed to handle competing objectives that change in time. The example shows how our approach allows to model each of the objectives separately and to combine them using game theory.