The increasing popularity of IoT devices allows us to communicate better, interact better, and ultimately build a new type of a scientific instrument that will allow us to explore our environment in ways that we could only dream about just a few years ago. This disruptive opportunity however raises a new set of challenges: how should we manage the massive amounts of data and network traffic such instruments will eventually produce? What types of environments will be most suited to developing their full potential? What new security problems will arise? And finally, what are the best ways of leveraging intelligent edge to create new types of applications?

In a research area that creates its own new reality, such questions are too often approached only theoretically for lack of a realistic testbed, a scientific instrument that keeps pace with the emergent requirements of science and allows researchers to deploy, measure, and analyze relevant scientific hypotheses. The NSF-funded Chameleon testbed, originally created to provide a platform for exploration of research topics in cloud computing -- such as design of new virtualization solutions, operating systems, or power management -- has now been extended to support experiments in cloud to edge.

In this talk, I will first describe the Chameleon testbed -- a scientific instrument for computer science systems research, originally created to allow exploration of research topics in cloud computing such as virtualization, programmable networking, or power management -- as well as its recent extensions to support experimentation at the edge. I will describe the testbed capabilities required to provide a platform for the edge to cloud continuum, and give examples of edge to cloud research and education projects our users are running. Finally, I will describe tools and methodologies that Chameleon provides to improve experimental methodology and reproducibility of experiments in this environment and illustrate how a common experimentation platform can enhance sharing and scientific productivity.

Given the increasingly pervasive role and growing importance of computing and data in all aspects of science and society fundamental advances in computer science and their translation to the real world have become essential. Consequently, there may be benefits to formalizing translational computer science (TCS) to complement the traditional foundational and applied modes of computer science research, as has been done for translational medicine. TCS has the potential to accelerate the impact of computer science research overall. In this talk I discuss the attributes of TCS, and formally define it. I enumerate a number of roadblocks that have limited its adoption to date and sketch a path forward. Finally, I will provide some specific examples of translational research in parallel and distributed computing, drawing on personal research and also of others in the field.

Over the past years, the use of Artificial Intelligence has become ubiquitous in most disciplines, and Scientific High Throughput applications are not the exception. Machine Learning increasingly plays a central role across the whole workflow pipeline, from workload forecasting, adaptive scheduling, self-managed resource allocation, and on the fly analysis. As we consider a pathway towards reproducible, scalable, and trustworthy science, we must pay special attention to the impact of ML in HTC, and how current practices can advance or hinder these efforts.

A common pitfall when designing ML-based solutions is the use of data "as is", and hoping that the models will automatically distill relevant features from the raw input. While this approach is feasible in domains with huge datasets, and no critical need for verifiable and reproducible results, this is a suboptimal approach in science, where data is expensive, very high dimensional, and noisy. This talk shows the power of embedding domain knowledge into the ML cycle, specifically within the data representation of complex entities, such as proteins. The goal of this encoding is to expose intra- and inter-molecular structural patterns to enable scalable and interpretable high throughput analyses. We present use cases in the context of protein function prediction and in-situ analysis of Molecular Dynamics simulations.