arXiv subjects

Chenyi Zhang

Publications and source records attributed to Chenyi Zhang.

45 records · Page 3Linked to original sources

arXiv2021

A Uniform Framework for Anomaly Detection in Deep Neural Networks

Deep neural networks (DNN) can achieve high performance when applied to In-Distribution (ID) data which come from the same distribution as the training set. When presented with anomaly inputs not from the ID, the outputs of a DNN should be regarded as meaningless. However, modern DNN often predict anomaly inputs as an ID class with high confidence, which is dangerous and misleading. In this work, we consider three classes of anomaly inputs, (1) natural inputs from a different distribution than the DNN is trained for, known as Out-of-Distribution (OOD) samples, (2) crafted inputs generated from ID by attackers, often known as adversarial (AD) samples, and (3) noise (NS) samples generated from meaningless data. We propose a framework that aims to detect all these anomalies for a pre-trained DNN. Unlike some of the existing works, our method does not require preprocessing of input data, nor is it dependent to any known OOD set or adversarial attack algorithm. Through extensive experiments over a variety of DNN models for the detection of aforementioned anomalies, we show that in most cases our method outperforms state-of-the-art anomaly detection methods in identifying all three classes of anomalies.

cs.LG↗

arXiv2021

Quantum algorithms for escaping from saddle points

We initiate the study of quantum algorithms for escaping from saddle points with provable guarantee. Given a function $f\colon\mathbb{R}^{n}\to\mathbb{R}$, our quantum algorithm outputs an $ε$-approximate second-order stationary point using $\tilde{O}(\log^{2} (n)/ε^{1.75})$ queries to the quantum evaluation oracle (i.e., the zeroth-order oracle). Compared to the classical state-of-the-art algorithm by Jin et al. with $\tilde{O}(\log^{6} (n)/ε^{1.75})$ queries to the gradient oracle (i.e., the first-order oracle), our quantum algorithm is polynomially better in terms of $\log n$ and matches its complexity in terms of $1/ε$. Technically, our main contribution is the idea of replacing the classical perturbations in gradient descent methods by simulating quantum wave equations, which constitutes the improvement in the quantum query complexity with $\log n$ factors for escaping from saddle points. We also show how to use a quantum gradient computation algorithm due to Jordan to replace the classical gradient queries by quantum evaluation queries with the same complexity. Finally, we also perform numerical experiments that support our theoretical findings.

quant-ph↗

arXiv2019

A Relational Static Semantics for Call Graph Construction

The problem of resolving virtual method and interface calls in object-oriented languages has been a long standing challenge to the program analysis community. The complexities are due to various reasons, such as increased levels of class inheritance and polymorphism in large programs. In this paper, we propose a new approach called type flow analysis that represent propagation of type information between program variables by a group of relations without the help of a heap abstraction. We prove that regarding the precision on reachability of class information to a variable, our method produces results equivalent to that one can derive from a points-to analysis. Moreover, in practice, our method consumes lower time and space usage, as supported by the experimental results.

cs.PL↗

arXiv2019

Characterising Probabilistic Alternating Simulation for Concurrent Games

Probabilistic game structures combine both nondeterminism and stochasticity, where players repeatedly take actions simultaneously to move to the next state of the concurrent game. Probabilistic alternating simulation is an important tool to compare the behaviour of different probabilistic game structures. In this paper, we present a sound and complete modal characterisation of this simulation relation by proposing a new logic based on probabilistic distributions. The logic enables a player to enforce a property in the next state or distribution. Its extension with fixpoints, which also characterises the simulation relation, can express a lot of interesting properties in practical applications.

cs.LO↗

arXiv2014

The Complexity of Synchronous Notions of Information Flow Security

The paper considers the complexity of verifying that a finite state system satisfies a number of definitions of information flow security. The systems model considered is one in which agents operate synchronously with awareness of the global clock. This enables timing based attacks to be captured, whereas previous work on this topic has dealt primarily with asynchronous systems. Versions of the notions of nondeducibility on inputs, nondeducibility on strategies, and an unwinding based notion are formulated for this model. All three notions are shown to be decidable, and their computational complexity is characterised.

cs.CR↗

arXiv2012

An Algorithm for Probabilistic Alternating Simulation

In probabilistic game structures, probabilistic alternating simulation (PA-simulation) relations preserve formulas defined in probabilistic alternating-time temporal logic with respect to the behaviour of a subset of players. We propose a partition based algorithm for computing the largest PA-simulation, which is to our knowledge the first such algorithm that works in polynomial time, by extending the generalised coarsest partition problem (GCPP) in a game-based setting with mixed strategies. The algorithm has higher complexities than those in the literature for non-probabilistic simulation and probabilistic simulation without mixed actions, but slightly improves the existing result for computing probabilistic simulation with respect to mixed actions.

cs.GT↗

arXiv2010

Unwinding Conditional Noninterference

Noninterference provides a control over information flow in a system for ensuring confidentiality and integrity properties. In the literature this notion has been well studied as transitive noninterference and intransitive noninterference. In this paper we define a framework on the notion of conditional noninterference, which allows to specify information flow policies based on the semantics of action channels. Our new policies subsume the policies of both transitive and intransitive noninterference, and support dynamic requirements such as upgrading and downgrading. We also present unwinding relations that are both sound and complete for the new policies.

cs.CR↗

arXiv2010

On Probabilistic Alternating Simulations

This paper presents simulation-based relations for probabilistic game structures. The first relation is called probabilistic alternating simulation, and the second called probabilistic alternating forward simulation, following the naming convention of Segala and Lynch. We study these relations with respect to the preservation of properties specified in probabilistic alternating-time temporal logic.

cs.LO↗

arXiv2009

How to Work with Honest but Curious Judges? (Preliminary Report)

The three-judges protocol, recently advocated by Mclver and Morgan as an example of stepwise refinement of security protocols, studies how to securely compute the majority function to reach a final verdict without revealing each individual judge's decision. We extend their protocol in two different ways for an arbitrary number of 2n+1 judges. The first generalisation is inherently centralised, in the sense that it requires a judge as a leader who collects information from others, computes the majority function, and announces the final result. A different approach can be obtained by slightly modifying the well-known dining cryptographers protocol, however it reveals the number of votes rather than the final verdict. We define a notion of conditional anonymity in order to analyse these two solutions. Both of them have been checked in the model checker MCMAS.

cs.CR↗