Search

Membership inference (MI) attacks highlight a privacy weakness in present stochastic training methods for neural networks. It is not well understood, however, why they arise. Are they a natural consequence of imperfect generalization only? Which underlying causes should we address during training to mitigate these attacks? Towards answering such questions, we propose the first approach to explain MI attacks and their connection to generalization based on principled causal reasoning. We offer causal graphs that quantitatively explain the observed MI attack performance achieved for $6$ attack variants. We refute several prior non-quantitative hypotheses that over-simplify or over-estimate the influence of underlying causes, thereby failing to capture the complex interplay between several factors. Our causal models also show a new connection between generalization and MI attacks via their shared causal factors. Our causal models have high predictive power ($0.90$), i.e., their analytical predictions match with observations in unseen experiments often, which makes analysis via them a pragmatic alternative., Comment: 26 pages, 15 figures; added CC-license block icons and links, typos corrected, added reference to Github

We design and implement from scratch a new fuzzer called SIVO that refines multiple stages of grey-box fuzzing. First, SIVO refines data-flow fuzzing in two ways: (a) it provides a new taint inference engine that requires only logarithmic in the input size number of tests to infer the dependency of all program branches on the input bytes, and (b) it deploys a novel method for inverting branches by solving directly and efficiently systems of inequalities. Second, our fuzzer refines accurate tracking and detection of code coverage with simple and easily implementable methods. Finally, SIVO refines selection of parameters and strategies by parameterizing all stages of fuzzing and then dynamically selecting optimal values during fuzzing. Thus the fuzzer can easily adapt to a target program and rapidly increase coverage. We compare our fuzzer to 11 other state-of-the-art grey-box fuzzers on 27 popular benchmarks. Our evaluation shows that SIVO scores the highest both in terms of code coverage and in terms of number of found vulnerabilities.

Automatic patch generation can significantly reduce the window of exposure after a vulnerability is disclosed. Towards this goal, a long-standing problem has been that of patch localization: to find a program point at which a patch can be synthesized. We present PatchLoc, one of the first systems which automatically identifies such a location in a vulnerable binary, given just one exploit, with high accuracy. PatchLoc does not make any assumptions about the availability of source code, test suites, or specialized knowledge of the vulnerability. PatchLoc pinpoints valid patch locations in large real-world applications with high accuracy for about 88% of 43 CVEs we study. These results stem from a novel approach to automatically synthesizing a test-suite which enables probabilistically ranking and effectively differentiating between candidate program patch locations.

The precise estimation of position is an essential concern for the control of a switched reluctance motor (SRM). Given the prominent role of position, the promising sensorless control approach for an SRM drive should be capable of providing accurate position. An inductance-based approach has been widely applied in the position estimation. However, the estimated accuracy suffers from the magnetic saturation effect, resulting in the poor performance of sensorless operation. This paper presents a reversible synchronization of commutation with the corresponding rotor position for SRM. With the consideration of magnetic saturation, the proposed approach is taken as a suitable candidate and plays an essential role in accommodating the requirement of optimal control. The relationship between a typical inductance position and magnetic circuit saturation is investigated. The instant of the intersections is sensed by comparing the instantaneous inductance of adjacent phases. Thus, the predicted position is obtained with the information of the special point and the calculated average speed. Compared with other existing methods, the proposed approach has the certain advantages, such as the ability to update the estimated speed and position six times per electrical period, which guarantees the estimated accuracy. The proposed approach is also valid even when the motor is operated at an acceleration state and heavy load operation. In addition, the requirement of the educated inductance structure is not unnecessary, less memory space is needed in the chip, and the accumulated error is eliminated. The simulation and experimental findings demonstrate the feasibility and practicality of the proposed position estimation approach with carrying out the inertial operation, load mutation, and high-speed test.

The high current ripple and torque ripple are the main drawbacks of the switched reluctance motor (SRM) since the nonlinearity and double saliency, which limits its applications. In order to eliminate the current variation and torque ripple, an optimized pulse width modulation (PWM) control is presented in this paper. The voltage ratio duty is able to be predicted precisely according to the information of the motor running parameter. Based on torque sharing functions (TSFs), the current profile is pre-computed and four regions are defined according to the reference current profiles. The three modes, excitation, demagnetization and freewheeling, are flexibly chosen according to the characteristic of the current profile. It is indicated that it is better than that of conventional PWM modulation in terms of current ripple and the current tracing performance is improved without increasing the switching frequency or changing the hysteresis band. The current ripple is defined as the peak-to-peak value dividing the average value and it is reduced by 40%. A comparison in terms of the torque ripple and copper loss is also carried out: the torque ripple is significantly reduced via the proposed scheme under both magnetic linear and saturation conditions. The torque ripple and copper loss are reduced by about 70% and 12%, respectively. The validity and effectiveness of the proposed control strategy is verified by simulation and experimental results.

Recently, neural models have been proposed for headline generation by learning to map documents to headlines with recurrent neural networks. Nevertheless, as traditional neural network utilizes maximum likelihood estimation for parameter optimization, it essentially constrains the expected training objective within word level rather than sentence level. Moreover, the performance of model prediction significantly relies on training data distribution. To overcome these drawbacks, we employ minimum risk training strategy in this paper, which directly optimizes model parameters in sentence level with respect to evaluation metrics and leads to significant improvements for headline generation. Experiment results show that our models outperforms state-of-the-art systems on both English and Chinese headline generation tasks.