2.4 KiB
2.4 KiB
type:: REVIEWS/PhDThesis
year:: 2023
full-title:: Automating Code-Related Tasks via Pre-trained MODELS of Code
tags:: people/Mastropaolo
external-links:: http://s.diruscio.org/2bu4a
date:: 25-01-2023 - 15:25
file::
- [[Highlights]]
- ((63d7cd11-e1df-4a63-b0da-e4e7a799c6ff))
- So this means that "most of the work" is done by the pretraining phase, thus they systems are already good as they are?
- Goals of the thesis
- **i) code snippet summarization**
- A lot of work has been done to document classes or methods. Little effort has been devoted to more fine-grained documenation (e.e., documenting code snippets or even single statement)
- **ii) generation and injection of complete log statements**
- This includes different research topics:
- Log message enhancement
- Log placement
- Log level recommendation
- Existing works address these aspects singularly, but none of them can provide a comprehensive support to developers. No complete generation and placement of complete log statements
- LANCE is the proposed approach published at ICSE2022 providing a complete support for synthesizing and injecting complete log statements in Java code (thus all the three sub-tasks above)
- Moreover
- Investigation of data-driven techniques to support the task of automated variable renaming in the context of Java methods
collapsed:: true
- ==why?==
- **Use of pre-trained language models to support the following core-related tasks:**
- Automatic bug-fixing
- Injection of code mutants
- Generation of assert statement in test methods
- Code summarization
- *One question is how such a list has been defined. All these tasks are indeed interesting, I'm just wondering why the student decided to pick those and focus their attention on them. Is it only depending on the available datasets?*
- That's the answer 🙂
collapsed:: true
- ((63da5db0-1788-49d0-981f-4f4a160c0096))
- ((63da5def-890e-49e0-ad12-5d30f45b5a21))
- Only one fine-tuning phase for all the tasks?
- It seems no, different scenarios have been implemented:
- ((63da605b-9f06-4373-8aab-ffd2ea7e3a9f))
- [[Comments]]
- 
-