Program

In the last two decades, different types of biometric measures have been adopted by researchers to complement traditional research in software engineering. Starting with eye-tracking, electro dermal activity (EDA), and Electroencephalogram (EEG) – just to name a few, researchers have even adopted brain imaging techniques in the last decade using functional Near Infrared Spectroscopy (fNIRS) and functional Magnetic Resonance Imaging (fMRI). Why are we doing this? What devices are used? What data is recorded? What measures are calculated? And how do we analyze, visualize, and make sense of that data? Those are the type of questions that we will address in this talk.

In this talk, we'll take an in-depth look at the award-winning study, 'First come first served: the impact of file position on code review.' The study investigates whether the order of files in code review tools can affect the review outcomes. This interactive talk will provide an extensive explanation of the research methods used in the study. We'll analyze each decision that shaped the study, the reasons behind these decisions, and the knowledge gained from this research approach. We'll explore the research paper's two-part process to demonstrate how the study's methods and their foundational principles can be applied to various empirical software engineering research scenarios. The goal of this session is to equip students with the hands-on experience necessary to plan, carry out, and evaluate their empirical studies, and help them approach real-world software development problems from a rigorous scientific perspective. First, we'll explain the two-step procedure that analyzed numerous Pull Requests from well-known Java projects on GitHub. We'll clarify how the data was gathered, managed, and understood. We'll also shed light on how to manage large-scale data effectively and control for variables that might distort the results. Then, we'll examine the design and implementation of a controlled experiment that involved 106 participants. We'll focus on elements like the selection of defects, experimental controls, assignment of treatments, and data analysis. We'll thoroughly discuss the process of transforming raw data into meaningful insights.

As we surf the crest of the latest AI tidal wave, we developers find ourselves savoring our portion of the technological banquet. Enthusiasm is boiling over, with shiny toys such as GitHub Copilot and ChatGPT heralding promises of skyrocketing our productivity. At CodeLounge, we have been using Copilot for a year, while also collecting data about its usage by means of Tako, an IDE telemetry collector & analyzer that we developed. In this talk, starting from this internal data, we will reflect on the impact that such tools had and might have on the structure and daily activity of a development team. Through various programming and data analysis tasks, we will try to shed some light into the line between promise and reality, connecting the dots to incisive critiques from leading researchers who question the capabilities of Large Language Models in a more general context. Can these parrots be our perfect pirate companions, finally taking up the cudgels against the tedium of our craft, like copy-pasting from Stack Overflow? Which strategies and best practices can be adopted as developers to do work, but also as team lead to structure the development process?

*: An anonymous stochastic parrot may have or may have not contributed to this abstract.

This talk will not give the answer to “The Ultimate Question of Life, The Universe, and Everything”. We already know the answer to that question; it is “42”.

However, this talk will answer several other questions that Ph.D. students may ask themselves at various stages of their doctoral studies. Some of these questions are:

• How to select a research topic and define a research agenda?
• What type of research is (un)common in software engineering?
• How should I conduct my research?
• Why should I (not) consider working in academia after graduation?

As we surf the crest of the latest AI tidal wave, we developers find ourselves savoring our portion of the technological banquet. Enthusiasm is boiling over, with shiny toys such as GitHub Copilot and ChatGPT heralding promises of skyrocketing our productivity. At CodeLounge, we have been using Copilot for a year, while also collecting data about its usage by means of Tako, an IDE telemetry collector & analyzer that we developed. In this talk, starting from this internal data, we will reflect on the impact that such tools had and might have on the structure and daily activity of a development team. Through various programming and data analysis tasks, we will try to shed some light into the line between promise and reality, connecting the dots to incisive critiques from leading researchers who question the capabilities of Large Language Models in a more general context. Can these parrots be our perfect pirate companions, finally taking up the cudgels against the tedium of our craft, like copy-pasting from Stack Overflow? Which strategies and best practices can be adopted as developers to do work, but also as team lead to structure the development process?

*: An anonymous stochastic parrot may have or may have not contributed to this abstract.

Together, we will take a deep dive and follow a less traveled path through the landscape of research methods in software engineering research. We will explore who our research aims to impact, what kinds of contributions we can expect from our research, and how we can use innovative research methods. Some of the topics we will dive into include design science as a frame for software engineering research, the benefits and challenges of using mixed methods in software engineering, and how to uncover the potential but not always obvious or positive disruptive impacts of novel technologies (such as generative AI and VR) on software engineering practice. After this talk, you should feel more empowered to pursue ambitious and impactful research using innovative research methods.

In recent years, Artificial Intelligence (AI) has become increasingly popular in the field of Software Engineering (SE), where it has been used to automate and improve various SE tasks. One of the most exciting developments in this area is the use of Large Language Models (LLMs) to solve complex SE problems. In this talk, we will explore the different techniques employed to use LLMs in AI for SE, including finetuning, prompt engineering, prompt augmentation with retrieval, and reinforcement learning with human feedback (RLHF) for SE tasks. Will examine the benefits and limitations of each approach and discuss real-world examples of their applications, such as in Automated Program Repair (APR). By the end of the talk, you will have a better understanding of how LLMs can be used in AI for SE and the potential impact they can have on the industry.

Software languages are more than their (in)formal specifications. They are also cultural artefacts that reflect the preferences and idioms of their users. In this session, we will explore some of the challenges and opportunities of seeing a language beyond its grammar. We will start by discussing how to determine the language of a program before parsing it, using various kinds of cues. Then, we will move on to capturing and analysing implicit rules and traditions that govern the style and structure of code, such as naming conventions, interfaces, idioms, implementation patterns, etc. Finally, we will address the issue of measuring and improving the quality of code in a specific language, taking into account its features and best practices. We will contemplate some examples of idiomatic and non-idiomatic code, and some methods for detecting and correcting them. The goal of this session is to demonstrate that seeing a language as a rich and diverse phenomenon can lead to new insights and applications for software engineering.