
Postdoctoral research associate @EdinburghUni, @EdinburghNLP, @EdiDataScience, @InfAtEd. Machine learning for software engineering.
Edinburgh, Scotland
Joined June 2012
- Tweets113
- Following194
- Followers116
- Likes266
Our new paper (with @RandomlyWalking ) 'Maybe Deep Neural Networks are the Best Choice for Modeling Source Code' is now available in arxiv: arxiv.org/abs/1903.05734
The Tensorflow implementation used for experiments is open source and available here: github.com/mast-group/OpenVo…
Rafael - Michael Karampatsis retweeted
We're giving away a limited-edition #Starfield @amdgaming GPU and CPU! Follow and repost for a chance to win - a winner will be messaged on August 30!
Rules: beth.games/3E664Ar
This Series X is armored up 🔥
RT with #XboxArmoredCoreSweepstakes & follow @Xbox and @armoredcore for a chance to win this ARMORED CORE VI FIRES OF RUBICON console and standard Series X bundle!
Ages 18+. NoPurchNec. Ends 9/1/23. Rules: bit.ly/3JADsTk

ALT Image of a custom ARMORED CORE VI FIRES OF RUBICON Xbox Series S, Xbox controller, and case in front of a dark sci-fi background. The Xbox logo is in the bottom left corner and the ARMORED CORE VI FIRES OF RUBICON logo is in the bottom right corner. Below the products reads, "©Bandai Namco Entertainment Inc. / ©1997-2023 FromSoftware, Inc. All rights reserved." The following line reads, "Game is not part of this sweepstakes. BUNDLE NOT FOR SALE".
Sharing exciting news with you! The ROG Gamescom 2023 online event will be live on August 22.
Tune in to the livestream to check out the latest product announcements, and enter the giveaway to win amazing ROG prizes! 🎉
Click the link to join! rog.gg/ROGGamescom23-twugc
Rafael - Michael Karampatsis retweeted
Sure, that's frustrating to us. But it's also hugely disrespectful to the reviewers/meta-reviewer who spend their valuable time, only to be completely disregarded. And what are we to do other than resubmit and waste more reviewer time? #AAAI2022 (2/2)
Rafael - Michael Karampatsis retweeted
Our @RealAAAI'22 *rejected* paper:
R1-Accept
R2-Weak Accept
R3-Borderline Accept
R4-Accept
Meta Review-"The reviewers all agree that this paper is above the acceptance bar. This paper has very interesting idea, proposes solid model, with thorough experiments and analysis."
(1/2)
This quoted post is unavailable.
Rafael - Michael Karampatsis retweeted
Excited to share🎺 our work on using #ML to automatically generate spreadsheet formulas 📈.
Main idea: Represent context from data in adjacent cells and header row using transformers 🤖
Spreadsheet formulas help users apply sophisticated analyses and powerful transformations to data, but writing formulas can be tedious and error-prone. Today we describe how #ML can make writing formulas easier by automatically generating them ↓ goo.gle/2Z5INh1
Rafael - Michael Karampatsis retweeted
Really interesting analysis!
"Inconsistency in Conference Peer Review: Revisiting the 2014 NeurIPS Experiment"
We logged the publication/citation status of over 1000 papers that were originally submitted to the 2014 conference.
Write up on ArXiv here: arxiv.org/abs/2109.09774
I entered for a chance to win an RTX Video Card from @jayztwocents Click this link for your chance too! gleam.io/liSGr/worldwide-rtx…
Rafael - Michael Karampatsis retweeted
"In-IDE Code Generation from Natural Language" by @frankxu2004, @b_vasilescu, @gneubig
📜 arxiv.org/abs/2101.11149
💻 github.com/neulab/tranX-plug…
IDE plugin for @pycharm + qualitative&qualitative evaluation of Code Synthesis impact on productivity of 31 developers at 14 tasks
Rafael - Michael Karampatsis retweeted
Pssst... Do you like automatic program repair? Empirical software engineering?... Here’s a fun shared challenge. Abstracts due JAN 19... Take a quick look at the data, I think you’ll like it!
Calling all Miners...Challenge awaits!
MSR21 features ManySStuBs4J, a dataset of fixes to Java simple bugs. @miltos1, Rafael-Michael Karampatsis and Charles Sutton
2021.msrconf.org/track/msr-2…
Rafael - Michael Karampatsis retweeted
Speaking of the msr challenge, @miltos1 from Microsoft Research, @MpatsisCS from Univ. Of Edinburgh, and @RandomlyWalking from GoogleAI are charing the msr challenge track.
Rafael - Michael Karampatsis retweeted
Small bugs are a fact of a programmer's life. But do they have to be? We invite you to participate in the Mining Software Repositories Challenge and study these bugs. There are many interesting research questions: Where & why these bugs appear? How can we detect & repair them?
Calling all Miners...Challenge awaits!
MSR21 features ManySStuBs4J, a dataset of fixes to Java simple bugs. @miltos1, Rafael-Michael Karampatsis and Charles Sutton
2021.msrconf.org/track/msr-2…
Rafael - Michael Karampatsis retweeted
🎉Congratulations to @EdiDataScience student @MpatsisCS for passing his viva this week! Well done!
Organizers of data science and machine learning conferences (NeurIPS, ICML, AISTATS, ICLR, UAI, ...): Allow remote paper & poster presentations at conferences - Sign the Petition! chng.it/S4gyndTc via @UKChange
Rafael - Michael Karampatsis retweeted
Our hearts go out to all those who may be battling COVID-19. Please help the #icse20 organization team to understand what the community feels should happen with #icse20 by answering a short survey: forms.gle/LyKXmTKse1JZGtpDA
Rafael - Michael Karampatsis retweeted
RT for a chance to win a Limited Edition Jordan Brand Custom Xbox One X console.
NoPurchNec. Ends 2/27. Rules: xbx.lv/38An6FL
Rafael - Michael Karampatsis retweeted
If you still have the energy, come to our poster on program synthesis with learned code idioms today 5-7pm, #168. Work by @rshin, strongly supervised by @alexpolozov and weakly supervised by @miltos1 and me. It's a fun trick to improve generation of things with common substructures.
Rafael - Michael Karampatsis retweeted
We also had great #DSRD19 talks from two of our @EdiDataScience students, @MichaelPJCamil and @MpatsisCS
Rafael - Michael Karampatsis retweeted
The CodeSearchNet Challenge 👩💻, a benchmark for evaluating #MLonCode methods by @HamelHusain @GitHubEng Includes:
- large parallel corpus of code & natural language (2M methods with descriptions);
- eval data annotated with query relevance by experts.
github.blog/2019-09-26-intro…




















