Software quality | EPFL Graph Search

In the context of software engineering, software quality refers to two related but distinct notions:

Software's functional quality reflects how well it complies with or conforms to a given design, based on functional requirements or specifications. That attribute can also be described as the fitness for purpose of a piece of software or how it compares to competitors in the marketplace as a worthwhile product. It is the degree to which the correct software was produced.
Software structural quality refers to how it meets non-functional requirements that support the delivery of the functional requirements, such as robustness or maintainability. It has a lot more to do with the degree to which the software works as needed.

Many aspects of structural quality can be evaluated only statically through the analysis of the software inner structure, its source code (see Software metrics), at the unit level, and at the system level (sometimes referred to as end-to-end testing), which is in

CauseOccam: Learning Interpretable Abstract Representations in Reinforcement Learning Environments via Model Sparsity

Sergei Volodin

"I choose this restaurant because they have vegan sandwiches" could be a typical explanation we would expect from a human. However, current Reinforcement Learning (RL) techniques are not able to provide such explanations, when trained on raw pixels. RL algorithms for state-of-the-art benchmark environments are based on neural networks, which lack interpretability, because of the very factor that makes them so versatile – they have many parameters and intermediate representations. Enforcing safety guarantees is important when deploying RL agents in the real world, and guarantees require interpretability of the agent. Humans use short explanations that capture only the essential parts and often contain few causes to explain an effect. In our thesis, we address the problem of making RL agents understandable by humans. In addition to the safety concerns, the quest to mimic human-like reasoning is of general scientific interest, as it sheds light on the easy problem of consciousness. The problem of providing interpretable and simple causal explanations of agent's behavior is connected to the problem of learning good state representations. If we lack such a representation, any reasoning algorithm's outputs would be useless for interpretability, since even the "referents" of the "thoughts" of such a system would be obscure to us. One way to define simplicity of causal explanations via the sparsity of the Causal Model that describes the environment: the causal graph has the fewest edges connecting causes to their effects. For example, a model for choosing the restaurant that only depends on the cause "vegan" is simpler and more interpretable than a model that looks at each pixel of a photo of the menu of a restaurant, and possibly relies as well on spurious correlations, such the style of the menu. In this thesis, we propose a framework "CauseOccam" for model-based Reinforcement Learning where the model is regularized for simplicity in terms of sparsity of the causal graph it corresponds to. The framework contains a learned mapping from observations to latent features, and a model predicting latent features at the next time-steps given ones from the current time-step. The latent features are regularized with the sparsity of the model, compared to a more traditional regularization on the features themselves, or via a hand-crafted interpretability loss. To achieve sparsity, we use discrete Bernoulli variables with gradient estimation, and to find the best parameters, we use the primal-dual constrained formulation to achieve a target model quality. The novelty of this work is in learning jointly a sparse causal graph and the representation taking pixels as the input on RL environments. We test this framework on benchmark environments with non-trivial high-dimensional dynamics and show that it can uncover the causal graph with the fewest edges in the latent space. We describe the implications of our work for developing priors enforcing interpretability.

2021

SQC: Secure Quality Control for Meta-Analysis of Genome-Wide Association Studies

Jacques Fellay, Zhicong Huang, Jean-Pierre Hubaux, Huang Lin

Motivation: Due to the limited power of small-scale genome-wide association studies (GWAS), researchers tend to collaborate and establish a larger consortium in order to perform large-scale GWAS. Genome-wide association meta-analysis (GWAMA) is a statistical tool that aims to synthesize results from multiple independent studies to increase the statistical power and reduce false-positive findings of GWAS. However, it has been demonstrated that the aggregate data of individual studies are subject to inference attacks, hence privacy concerns arise when researchers share study data in GWAMA. Results: In this article, we propose a secure quality control (SQC) protocol, which enables checking the quality of data in a privacy-preserving way without revealing sensitive information to a potential adversary. SQC employs state-of-the-art cryptographic and statistical techniques for privacy protection. We implement the solution in a meta-analysis pipeline with real data to demonstrate the efficiency and scalability on commodity machines. The distributed execution of SQC on a cluster of 128 cores for one million genetic variants takes less than one hour, which is a modest cost considering the 10-month time span usually observed for the completion of the QC procedure that includes timing of logistics.

Oxford University Press2017

On Secure Cloud Computing for Genomic Data: From Storage to Analysis

Zhicong Huang

Although privacy is generally considered to be the right of an individual or group to control information about themselves, such a right has become challenging to protect in the digital era, this is exemplified by the case of cloud-based genomic computing. Despite the rapid progress in understanding, producing, and using genomic information, the practice of genomic data protection remains a fairly underdeveloped area. One of the indisputable reasons is that most nonexpert individuals do not realize the sensitive nature of their genomic data, unless it has been used against them. Many commercial organizations take advantage of their customers by taking control of personal genomic information, if customers want to benefit from services such as genetic analysis; even worse, these organizations often do not enforce proper protection, which could result in embarrassing data breaches. In this thesis, we investigate the potential threats of cloud- based genomic computing systems and propose various countermeasures by taking into account the functionality requirement. We begin with the most basic system where only symmetric encryption is needed for the cloud storage of genomic data, and we propose a new solution that protects the data against brute-force attacks that threaten the security of password-based encryption in direct-to-consumer companies. The solution employs honey encryption, where plaintext messages need to be transformed to a different space with uniform distribution on elements. We present a novel distribution-transformation encoder. We provide formal security proof of our solution. We analyze the scenario where efficient searching on encrypted data is necessary. We propose a system that provides fast retrieval on encrypted compressed data and that enables individuals to authorize access to fine-grained regions during data retrieval. Our solution addresses three critical dimensions in platforms that use large genomic data: encryption, compression, and efficient data retrieval. Compared with a previous de facto standard solution for storing aligned genomic data, our solution uses 18% less storage. To enable complicated data analysis, we focus on a proposal for secure quality-control of genomic data by using secure multi-party computation based on garbled circuits. Our proposal is for aggregated genomic data sharing, where researchers want to collaborate to perform large-scale genome-wide association studies in order to identify significant genetic variants for certain diseases. Data quality control is the very first stage of such a collaboration and remains a driving factor for further steps. We investigate the feasibility of advanced cryptographic techniques in the data protection of this phase. We demonstrate that for certain protocols, our solution is efficient and scalable. With the advent of precision medicine based on genomic data, the future of big data has become clearly inseparable from cloud-based genomic computing. It is important to continuously re-evaluate the standards of cloud-based genomic computing as novel technologies are developed, security threats arise, and more complex genomic analyses become possible. This is not only a battle against cyber criminals, but also against rigid and ignorant practices. Integrative solutions that carefully consider the use and misuse of personal genomic data are essential for ensuring secure, effective storage and maximizing utility in treating and preventing disease.