Colloquium (C)

The research field of Natural Language Processing (NLP) puts a lot of emphasis on empirical results. It seems that models are reaching state-of-the-art and even “super-human” performance on language understanding tasks on a daily basis, thanks to large datasets and powerful models with billions of parameters. However, the existing evaluation methodologies are lacking in rigor, leaving the field susceptible to erroneous claims. In this talk, I will describe efforts to build a solid framework for evaluation and experiment design for diverse NLP tasks.

Machine learning (ML) is integrated into our society, it is present in the judicial, health,  transportation, and financial systems. As the integration increases, the necessity of ML transparency increases. The fields of explainable and interpretable ML attempt to add transparency to ML: either by adding explanations to a given black-box ML model or by building a model which is interpretable and self-explanatory.

In data science, it is increasingly the case that the main challenge is finding, curating, and understanding the data that is available to solve a problem at hand. Furthermore, modern-day data is challenging in that it lacks many forms of semantics ("meaning of data"). Metadata may be incomplete or unreliable, data sources are unknown, and data documentation rarely exists. To address these challenges, the objective of my research is to recover data semantics throughout data discovery, versioning, integration, and quality.

In recent years, finding a path for a moving agent has become relevant to many real-world applications, including autonomous driving cars, moving robots, and flying drones. Finding the optimal (shortest) path can be calculated by Heuristic Search algorithms, such as the well-known A* algorithm. These algorithms use heuristics, which estimate the distance to the destination, to intelligently guide the search and quickly return a solution. In automated warehouses and many other applications, multiple agents exist.

Natural Language Processing (NLP) is a branch of Artificial Intelligence that aims to build machines that humans can seamlessly interact with through spoken and written language. As NLP becomes more ubiquitous in our daily lives, through technologies like Google Translate and Apple’s Siri, understanding the limits of these systems is critical. In the first half of the talk, we will cover my work developing diagnostics that test such systems.

Recently, there has been a growing interest in harnessing the power of big datasets and modern machine learning for designing new scalable algorithms. This invites us to rethink the role of data in algorithm design: not just as the input to pre-designed algorithms, but also a factor that enters the algorithm design process itself, driving it in a strong and possibly automated manner.

Therapy response varies considerably among cancer patients and depends on tumor intrinsic factors and interactions with the tumor microenvironment and the host immune system. One example is the response of colorectal cancer (CRC) patients to immune therapy, which varies considerably between patients with mismatch repair-deficient (MMRd) and mismatch-repair proficient tumors (MMRp).

In this talk I will survey recent developments in approximate subgraph-counting and subgraph-sampling in sublinear-time. Counting and sampling small subgraphs (aka motifs) are two of the most basic primitives in graph analysis, and have been studied extensively, both in theory and in practice. In my talk, I will present the sublinear-time computational model, where access to the input graph is given only through queries, and will explain some of the concepts that underlie my results.

The modern economy is becoming highly digital, demanding a combination of both economic and computational techniques. The abundance of data and the unique nature of digital markets highlight the special role of information in today’s economic systems. In this talk, I will discuss two domains of interest. The first is auction design in the interdependent value (IDV) setting. In IDV settings, buyers hold partial information regarding the quality of the good being sold, but their actual values also depend on information that other buyers have about the good.

Data science has become prevalent in various fields that affect day-to-day lives, such as healthcare, banking, and the job market. The process of developing data science applications usually consists of several automatic systems that manipulate and prepare the data in different manners. Examples of automatic data manipulations and preparations include generating synthetic data, exploring it, repairing the data, and labeling it for machine learning.