Colloquium: Dr. Kush Varshney, IBM, Trustworthy Machine Learning
Title
Trustworthy Machine Learning
Trustworthy Machine Learning
Resource management significantly impacts the performance and availability of cloud infrastructure. In the first part of my talk, I will introduce my work on managing memory, cache, and power resources to boost the performance and availability of cloud infrastructure. On memory resources, we propose UH-MEM, a utility-based data placement technique for hybrid memory.
Algorithm designers often introduce random choices into their algorithms in an effort to improve efficiency. However, random bits cannot necessarily be produced for free, so deterministic algorithms are preferable to randomized algorithms, all else being equal. Is randomness ever truly necessary for efficient computation?
What, ultimately, is the role of randomness in computing?
The large majority of existing pieces of software in industry are long-lived systems (a.k.a., legacy systems) usually developed using a monolithic architecture. But, over the years, user requirements changed, technologies evolved, and new business models emerged, leading to changes of such systems. As a result of extensive maintenance and obsolete technology, legacy systems usually have decayed and degraded architectures.
Sources of big data are now near-universal, e.g., across social media, financial markets, Internet of things, or scientific data. Thus understanding how to effectively and efficiently summarize data is a crucial step toward both understanding past events as well as predicting
Machine Learning is rapidly revolutionizing the development of many modern-day systems. However, testing Machine Learning-based systems is challenging due to
In this talk, I will present an algorithmic foundation of parallel paging and an overview of our recent results on performing general linear stencil computations significantly faster than state-of-the-art algorithms. Classical problems such as paging have been very well understood in the sequential setting for decades. However, the paging problem has remained wide open for more than two decades in the parallel setting. In the parallel paging problem, p processors share a cache (small, fast memory) of size k.
Abstractions have proven essential for us to manage computing systems that are constantly growing in size and complexity. However, as core design primitives are
Machine Learning (ML) techniques have been increasing adopted for Software Engineering (SE) tasks, such as code completion and code summarization. However, existing ML models provide limited value for SE tasks, because these models do not take into account the key characteristics of software: software is executable and software constantly evolves. In this talk, I will present my insights and work on developing execution-guided and evolutionaware ML models for several SE tasks targeting important domains, including software testing, verification, and maintenance.
Technologies that enable confidential communication and anonymous authentication are important for improving privacy for users of internet services. Unfortunately, encryption and anonymity, while good for privacy, make it hard to hold bad actors accountable for misbehavior. Internet services rely on seeing message content to detect spam and other harmful content; services must also be able to identify users to attribute and respond to abuse complaints.