Search

Supplementary Material 1

Additional file 1. Supplementary figures and tables.

Building advanced machine learning (ML) models requires expert knowledge and many trials to discover the best architecture and hyperparameter settings. Previous work demonstrates that model information can be leveraged to assist other attacks, such as membership inference, generating adversarial examples. Therefore, such information, e.g., hyperparameters, should be kept confidential. It is well known that an adversary can leverage a target ML model's output to steal the model's information. In this paper, we discover a new side channel for model information stealing attacks, i.e., models' scientific plots which are extensively used to demonstrate model performance and are easily accessible. Our attack is simple and straightforward. We leverage the shadow model training techniques to generate training data for the attack model which is essentially an image classifier. Extensive evaluation on three benchmark datasets shows that our proposed attack can effectively infer the architecture/hyperparameters of image classifiers based on convolutional neural network (CNN) given the scientific plot generated from it. We also reveal that the attack's success is mainly caused by the shape of the scientific plots, and further demonstrate that the attacks are robust in various scenarios. Given the simplicity and effectiveness of the attack method, our study indicates scientific plots indeed constitute a valid side channel for model information stealing attacks. To mitigate the attacks, we propose several defense mechanisms that can reduce the original attacks' accuracy while maintaining the plot utility. However, such defenses can still be bypassed by adaptive attacks., Comment: To appear in the 32nd USENIX Security Symposium, August 2023, Anaheim, CA, USA

Ni-rich layered oxides are promising cathodes to satisfy the long driving range of electric vehicles. However, Li/Ni mixing becomes a critical issue, which affects nearly all of the aspects of elec...

Firstly, compared with the traditional vehicle interactive verification methods, it is concluded that the real vehicle test scheme has the disadvantages of high cost, poor experience of pure physical bench scheme and lack of force feedback of pure virtual reality scheme. Augmented reality plus physical bench scheme has potential. Through the disassembly of automobile interactive verification process, the factors affecting the effect of automobile interactive verification are summarized. Secondly, conduct investigation and interview on the automobile active interactive users and the subjects of verification experiment, summarize the form and scope of automobile human-computer interaction design, as well as the design strategies and difficulties of interactive verification experiment, so as to determine the functional requirements of automobile interactive design verification bench. Then the feasibility of augmented reality technology applied to automotive human-computer interaction design verification is discussed and practiced.

The progress of science and technology and the continuous development of society let us march towards an intelligent society, and the vehicle interaction mode will change accordingly. This paper aims to analyze the development of the interaction mode of the vehicle and the influencing factors of the user's travel experience, to predict the interaction mode of intelligent vehicles in the context of smart cities, and to provide some help for the relevant forward-looking design. The final conclusion is that the interaction research of smart vehicles should continue to focus on both in-vehicle interaction and out-of-vehicle interaction. The positioning of vehicles has changed: from travel tools to intelligent space development, interaction mode from two-dimensional plane to three-dimensional space development, there is a traditional single task of a single interaction mode to multi-scene co-existence, multi-modal interaction development. Users' social, entertainment and personalization needs in the vehicle will influence the overall travel experience of the user.

Differential steering is a unique steering technology for distributed drive vehicles, which cannot only be applied to steering power, but also be used as a backup steering scheme for distributed drive vehicles. When the road adhesion conditions are poor, differential steering will lead to wheel slip, reduce the tire lateral force margin, and then affect the vehicle stability. To solve this issue, it is necessary to integrate the differential steering and anti-skid drive control. In this paper, the four-wheel distributed drive electric vehicle (DDEV), tire and wheel dynamics models are firstly established. Then the coordinated control strategy is proposed for anti-skid driving and differential steering of DDEV, where a neural network is adopted for the weights regulation of the two controllers. Next, an improved slip rate estimation method avoiding the calculation of vehicle speed is proposed which can reduce the error when the wheel angular acceleration is less than zero. Then, the slip rate fuzzy threshold controller is designed to control the wheel slip rate, and the feed forward-feedback differential torque controller is designed to obtain the differential torque required in steering. Considering the vehicle stability, the torque distribution adopts the quadratic programing method and combines the constraints of wheel slip rate and tire load rate to optimize the distribution of each wheel’s driving torque. Finally, the joint simulation and hardware-in-the-loop (HIL) are carried out to verify the effectiveness of the proposed improved slip rate estimation method and control strategy. The results show that the calculation error is reduced to 5.13% with the improved slip rate estimation results, and the superiority of proposed control method is verified by both simulation and HIL test.

Distributed drive vehicles are prone to wheel slip during driving on low adhesion coefficient roads. Wheel slip will not only cause energy loss, but also the different driving states of the two sides of the wheels will lead to a sharp deterioration in vehicle stability, which will adversely affect the dynamics and safety of the vehicle. For the control characteristics of distributed drive vehicles, a slip rate controller is designed on the basis of slip rate estimation, and a stability control strategy adapted to straight and steering driving are proposed. Firstly, the road surface is identified based on the Burckhardt tire model, and the optimal slip rate of the current road surface is estimated. And the optimal wheel speed corresponding to the current vehicle speed is calculated. An active disturbance rejection controller (ADRC) is established, which controls the four-wheel speeds by adjusting the motor output torque and tracks the optimal wheel speed corresponding to the optimal slip rate. The sliding-mode controller is designed considering the stability requirements of the vehicle during high-speed steering. And the wheel output torque is optimally allocated based on the quadratic programing method. Finally, joint simulations and hardware-in-the-loop tests verify the effectiveness of the control strategy proposed in this paper.

In order to know the travel dilemma of the elderly in the old mountain city area where the aged people are gathered but full of contradictions in life under the background of rapid and deep aging in China, this study summarizes the characteristics and requirements of the elderly travel by using such methods as literature review, field investigation and hierarchical analysis, extracts the relevant evaluation factors of traffic aging and determines their weights, and constructs the evaluation system of the suitability of transplantation to provide scientific and powerful ideas and basis for urban traffic planning for helping the elderly in the future.

With the rapid development of AI technologies, thousands of AI papers are being published each year. Many of these papers have released sample code to facilitate follow-up researchers. This paper presents an explorative study of over 1700 code repositories of AI papers hosted on GitHub. We find that these repositories are often poorly written, lack of documents, lack of maintenance, and hard to configure the underlying runtime environment. Thus, many code repositories become inactive and abandoned. Such a situation makes follow-up researchers hard to reproduce the results or do further research. In addition, these hard-to-reuse code makes a gap between academia and industry. Based on the findings, we give some recommendations on how to improve the quality of code repositories of AI papers.

Based on the example of key national heritage conservation unit---the protection and utilization planning of Fengxiangxi Conference Site in Dejiang County, Guizhou Province, this paper explores the effective measures for the protection and utilization planning of the revolutionary site from a multi-disciplinary perspective by expounding the theme of its planning and utilization, the delimitation of the scope of protection and the maintenance of building units.

Fibroblast growth factor 21 (FGF21), liver-derived hormone, exerts diverse metabolic effects, being considered for clinical application to treat obesity and diabetes. However, its anorexigenic effect is debatable and whether it involves the central mechanism remains unclarified. Moreover, the neuron mediating FGF21’s anorexigenic effect and the systemic energy state supporting it are unclear. We explored the target neuron and fed/fasted state dependence of FGF21’s anorexigenic action. Intracerebroventricular (ICV) injection of FGF21 markedly suppressed food intake in fed mice with elevated blood glucose. FGF21 induced c-Fos expression preferentially in hypothalamic paraventricular nucleus (PVN), and increased mRNA expression selectively for nucleobindin 2/nesfatin-1 (NUCB2/Nesf-1). FGF21 at elevated glucose increased [Ca2+]i in PVN NUCB2/Nesf-1 neurons. FGF21 failed to suppress food intake in PVN-preferential Sim1-Nucb2-KO mice. These findings reveal that FGF21, assisted by elevated glucose, activates PVN NUCB2/Nesf-1 neurons to suppress feeding under fed states, serving as the glycemia-monitoring messenger of liver-hypothalamic network for integrative regulation of energy and glucose metabolism.

Nitrogen and phosphorus adsorbents are widely used to mitigate agricultural non-point source pollution. However, research on adsorbents mainly involves studying chemical adsorption properties, and analyzes of the effects of adsorbent on pollutant removal has not considered the surface morphology of the adsorbent or the surface distribution of pollutants. In this study, we focus on the surface morphology of the adsorbent and the surface distribution of contaminants while examining chemical adsorption properties. The crystal composition of the adsorbent was evaluated by x-ray diffraction (XRD) characterization. Kinetic adsorption data and adsorption isotherms demonstrated that thermally modified zeolite exhibits better nitrogen adsorption. The optimal removal of nitrogen and phosphorus by thermally modified zeolite and diatomite occurred at a 3:2 ratio, reaching a removal rate of 92.07% and 84.61%, respectively. The potential adsorption mechanism of a composite adsorbent for nitrogen and phosphorus capture was investigated by Fourier transform infrared spectroscopy. Scanning electron microscopy mapping, grey image recognition, and gradient recognition confirmed a relationship between the surface morphology of the adsorbent and the distribution of surface pollutants. The larger the surface of the gradient, the more uneven it is, the more nitrogen and phosphorus sites are adsorbed on the surface, and the more nitrogen and phosphorus are adsorbed. These results suggest that thermally modified zeolite/diatomite can serve as a promising adsorbent for nitrogen and phosphorus removal in practical applications.

The present study demonstrated that the chitinase gene ChiKJ406136 of Streptomyces sampsonii (Millard & Burr) Waksman KJ40 could be cloned using a PCR protocol and expressed in Escherichia coli (Migula) Castellani & Chalmers BL21 (DE3), and the recombinant protein had antifungal effect on four forest pathogens (Cylindrocladium scoparium Morgan, Cryphonectria parasitica (Murrill) Barr, Neofusicoccum parvum Crous, and Fusarium oxysporum Schl.) and also had the biological control effects on Eucalyptus robusta Smith leaf blight, Castanea mollissima BL. blight, Juglans regia L. blight and J. regia root rot. The results showed that ChiKJ406136 was efficiently expressed and a 48 kilodalton (kDa) recombinant protein was obtained. No significant change in protein production was observed in the presence of different concentrations of IPTG (isopropyl-b-D-thio-galactoside). The purified protein yield was greatest in the 150 mmol/L imidazole elution fraction, and the chitinase activities of the crude protein and purified protein solutions were 0.045 and 0.033 U/mL, respectively. The antifungal effects indicated that mycelial cells of the four fungi were disrupted, and the control effects of the chitinase on four forest diseases showed significant differences among the undiluted 10- and 20-fold dilutions and the control. The undiluted solution exhibited best effect. The results of this study provide a foundation for the use of S. sampsonii as a biocontrol agent and provides a new source for the chitinase gene, providing a theoretical basis for its application.

The paper develops a simple model to demonstrate that, paradoxically, greater competition may exacerbate the problem of corruption. Market participants engaging in corrupt practices enjoy lower production costs -- maybe because they pay a bribe to avoid installing the environmental safeguards required by law -- such that honest players are driven out of the market when the market becomes sufficiently competitive.

The surface topography plays an important role in surface quality of cold-rolled steel strip.How surface topography changed in temper rolling is investigated.Based on Ritz method,a new numerical solution of steel strip plastic deformation during temper rolling is developed.Combining it with elastic deformation of rolls,the integration model of rolls and strip is built.Then the elastic-plastic contact between work roll and steel strip is simplified to asperity contact between one rough surface and one ideal smooth surface.The micro models of the generation of steel strip surface topography with pressing-in,furrowing and squeezing are built.The models considered the uplift effect of edges and neighbor role of asperities.Combining two kinds of models mentioned above,a model for the generation of surface topography is established considering asperity contact,roll force,relative sliding displacement between work roll and steel strip.The generation of surface topography of high strength steel strip and the influence of process parameters during temper rolling are studied.The results show that the pressing-in,furrowing and squeezing of work roll asperities on steel strip all influenced the strip surface morphology.The surface roughness of steel strip in the outlet with the changes in tension stress,reduction,roll roughness,roughness of steel strip at the entrance are obtained when pressing-in,furrowing and squeezing are dominant respectively.The following experiments validated the calculation results.