Computer vision and deep neural networks have been significantly promoting the development of visual perception in these years. Particularly, for autonomous vehicles, real-time image/video data is captured by onboard cameras and analyzed by computer vision techniques in many real applications. In the captured camera data, some contents can be used as auxiliary information to infer individuals’ locations and trajectories, which leads to severe privacy leakage but has been rarely studied. Thus, the goal of this article is to protect individuals’ location privacy by hiding side-channel information in the captured data while preserving the data utility for downstream applications. To this end, the technology of generative adversarial networks (GAN) is utilized to design two novel models, named ADGAN-I and ADGAN-II, both of which can take the original camera data as inputs and generate privacy-preserving outputs according to predefined sensitive object class. Thus, the processed camera data can defend location inference attack from adversaries in offline applications. Moreover, in ADGAN-I and ADGAN-II, the tradeoff between location privacy and data utility can be effectively balanced. Finally, the results of extensive real-data experiments validate the superiority of our proposed models over the state of the arts in utility preservation and privacy protection for autonomous vehicles’ images and videos.

ADGAN: Protect Your Location Privacy in Camera Data of Auto-Driving Vehicles

Accessibility of digital libraries and other web-based repositories has caused the illusion of accessibility of the full texts of scientific papers. However, in the majority of cases such an access (at least free access) is limited only to abstracts having no more then 50-100 words. Traditional keyword-based approach for clustering this type of documents gives unstable and imprecise results. We show that they can be easy improved with more adequate keyword selection and document similarity evaluation. We suggest simple procedures for this. We evaluate our approach on the data from two international conferences. One of our conclusions is the suggestion for the digital libraries and other repositories to provide document images of full texts of the papers along with their abstracts for open access via Internet.

Clustering abstracts instead of full texts

In order to solve the problem of human motion recognition in multimedia interaction scenarios in virtual reality environment, a motion classification and recognition algorithm based on linear decision and support vector machine (SVM) is proposed. Firstly, the kernel function is introduced into the linear discriminant analysis for nonlinear projection to map the training samples into a high-dimensional subspace to obtain the best classification feature vector, which effectively solves the nonlinear problem and expands the sample difference. The genetic algorithm is used to realize the parameter search optimization of SVM, which makes full use of the advantages of genetic algorithm in multi-dimensional space optimization. The test results show that compared with other classification recognition algorithms, the proposed method has a good classification effect on multiple performance indicators of human motion recognition and has higher recognition accuracy and better robustness.

/pdf/human-motion-recognition-based-on-svm-in-vr-art-media-tx6g54mr6y.pdf

Human motion recognition based on SVM in VR art media interaction environment

Deep neural networks (DNNs) have demonstrated their superiority in practice. Arguably, the rapid development of DNNs is largely benefited from high-quality (open-sourced) datasets, based on which researchers and developers can easily evaluate and improve their learning methods. Since the data collection is usually time-consuming or even expensive, how to protect their copyrights is of great significance and worth further exploration. In this paper, we revisit dataset ownership verification. We find that existing verification methods introduced new security risks in DNNs trained on the protected dataset, due to the targeted nature of poison-only backdoor watermarks. To alleviate this problem, in this work, we explore the untargeted backdoor watermarking scheme, where the abnormal model behaviors are not deterministic. Specifically, we introduce two dispersibilities and prove their correlation, based on which we design the untargeted backdoor watermark under both poisoned-label and clean-label settings. We also discuss how to use the proposed untargeted backdoor watermark for dataset ownership verification. Experiments on benchmark datasets verify the effectiveness of our methods and their resistance to existing backdoor defenses. Our codes are available at \url{https://github.com/THUYimingLi/Untargeted_Backdoor_Watermark}.

Untargeted Backdoor Watermark: Towards Harmless and Stealthy Dataset Copyright Protection

ICT-trade and pollution in Africa: Do governance and regulation matter?

Clustering narrow-domain short texts, such as academic abstracts, is an extremely difficult clustering problem. Firstly, short texts lead to low frequency and sparseness of words, making clustering results highly unstable and inaccurate; Secondly, narrow domain leads to great overlapping of insignificant words and makes it hard to distinguish between sub-domains, or fine-grained clusters. The vocabulary size is also too small to construct a good word bag needed by traditional clustering algorithms like LDA to give a meaningful topic distribution. A novel clustering model, Partitioned Word2Vec-LDA (PW-LDA), is proposed in this paper to tackle the described problems. Since the purpose sentences of an abstract contain crucial information about the topic of the paper, we firstly implement a novel algorithm to extract them from the abstracts according to its structural features. Then high-frequency words are removed from those purpose sentences to get a purified-purpose corpus and LDA and Word2Vec models are trained. After combining the results of both models, we can cluster the abstracts more precisely. Our model uses abstract text instead of keywords to cluster because keywords may be ambiguous and cause unsatisfied clustering results shown by previous work. Experimental results show that the clustering results of PW-LDA are much more accurate and stable than state-of-the-art techniques.

LDA Meets Word2Vec: A Novel Model for Academic Abstract Clustering

Compared with tedious per-pixel mask annotating, it is much easier to annotate data by clicks, which costs only several seconds for an image. However, applying clicks to learn video semantic segmentation model has not been explored before. In this work, we propose an effective weakly-supervised video semantic segmentation pipeline with click annotations, called WeClick, for saving laborious annotating effort by segmenting an instance of the semantic class with only a single click. Since detailed semantic information is not captured by clicks, directly training with click labels leads to poor segmentation predictions. To mitigate this problem, we design a novel memory flow knowledge distillation strategy to exploit temporal information (named memory flow) in abundant unlabeled video frames, by distilling the neighboring predictions to the target frame via estimated motion. Moreover, we adopt vanilla knowledge distillation for model compression. In this case, WeClick learns compact video semantic segmentation models with the low-cost click annotations during the training phase yet achieves real-time and accurate models during the inference period. Experimental results on Cityscapes and Camvid show that WeClick outperforms the state-of-the-art methods, increases performance by 10.24% mIoU than baseline, and achieves real-time execution.

https://dl.acm.org/doi/pdf/10.1145/3474085.3475217

WeClick: Weakly-Supervised Video Semantic Segmentation with Click Annotations

Privacy protection is an important research area, which is especially critical in this big data era. To a large extent, the privacy of visual classification data is mainly in the mapping between the image and its corresponding label, since this relation provides a great amount of information and can be used in other scenarios. In this paper, we propose the mapping distortion based protection (MDP) and its augmentation-based extension (AugMDP) to protect the data privacy by modifying the original dataset. In the modified dataset generated by MDP, the image and its label are not consistent ($e.g.$, a cat-like image is labeled as the dog), whereas the DNNs trained on it can still achieve good performance on benign testing set. As such, this method can protect privacy when the dataset is leaked. Extensive experiments are conducted, which verify the effectiveness and feasibility of our method. The code for reproducing main results is available at \url{this https URL}.

Visual Privacy Protection via Mapping Distortion

Designing proper loss functions for vision tasks has been a long-standing research direction to advance the capability of existing models For object detection, the well-established classification and regression loss functions have been carefully designed by considering diverse learning challenges (eg class imbalance, hard negative samples, and scale variances) Inspired by the recent progress in network architecture search, it is interesting to explore the possibility of discovering new loss function formulations via directly searching the primitive operation combinations So that the learned losses not only fit for diverse object detection challenges to alleviate huge human efforts, but also have better alignment with evaluation metric and good mathematical convergence property Beyond the previous auto-loss works on face recognition and image classification, our work makes the first attempt to discover new loss functions for the challenging object detection from primitive operation levels and finds the searched losses are insightful We propose an effective convergence-simulation driven evolutionary search algorithm, called CSE-Autoloss, for speeding up the search progress by regularizing the mathematical rationality of loss candidates via two progressive convergence simulation modules: convergence property verification and model optimization simulation CSE-Autoloss involves the search space (ie 21 mathematical operators, 3 constant-type inputs, and 3 variable-type inputs) that cover a wide range of the possible variants of existing losses and discovers best-searched loss function combination within a short time (around 15 wall-clock days with 20x speedup in comparison to the vanilla evolutionary algorithm) We conduct extensive evaluations of loss function search on popular detectors and validate the good generalization capability of searched losses across diverse architectures and various datasets Our experiments show that the best-discovered loss function combinations outperform default combinations (Cross-entropy/Focal loss for classification and L1 loss for regression) by 11% and 08% in terms of mAP for two-stage and one-stage detectors on COCO respectively Our searched losses are available at https://githubcom/PerdonLiu/CSE-Autoloss

/pdf/loss-function-discovery-for-object-detection-via-convergence-2252zcxb3j.pdf

Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search

More countries have made carbon neutral or net zero emission commitments since 2019. Within this context, re-examining the environmental Kuznets curve (EKC) hypothesis plays an essential role in sizing up the global economic development situation and realizing the global carbon emission reduction target. A methodological challenge in testing the EKC hypothesis, which states that increasing income makes CO2 emissions begin to decline beyond a turning point, lies in determining if this benchmark point exists. The EKC hypothesis between income and CO2 emissions is reassessed by applying a new kink regression model for the G7 countries from 1890 to 2015. Results reveal the inverted U-shaped nexus does not exist for US, Germany, Italy, Canada and Japan. For these five countries, the EKC curve has a turning point, but the positive impact of incomes on CO2 emissions becomes significantly smaller after the turning point. We describe this relationship as a pseudo-EKC. K.U.K. and France are the only exceptions, fitting the EKC hypothesis. Further analysis indicates that the relationship between income and SO2 emissions presents an inverted U-shaped curve. Moreover, we observe that the turning point occurs at different points in time for the different G7 countries. Therefore, environmental policies targeting pollutant emission reduction should consider the different characteristics of different pollutants and regions.

/pdf/re-examining-the-income-co2-emissions-nexus-using-the-new-1i3s5dms.pdf

Peidong Liu

Papers

LDA Meets Word2Vec: A Novel Model for Academic Abstract Clustering

WeClick: Weakly-Supervised Video Semantic Segmentation with Click Annotations

Visual Privacy Protection via Mapping Distortion

Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search

Re-Examining the Income–CO2 Emissions Nexus Using the New Kink Regression Model: Does the Kuznets Curve Exist in G7 Countries?