Miltos Allamanis 🇪🇺

@miltos1

Researching deep learning for generating and understanding programs. Research Scientist @GoogleAI

Also at @miltos@sigmoid.social

(Opinions are my own.)

miltos.allamanis.com Joined March 2008

321 Following 1.3K Followers 239 Posts

Latest posts 5

Miltos Allamanis 🇪🇺 @miltos1 Dec 27, 2022

This is an interesting result! But does "filtering-for-stars" tells us something about the included code quality/quanitity or the benchmark (HumanEval & co)? (probably something about both...)

BigCode @BigCodeProject Dec 22, 2022

In addition to the standard near-deduplication and heuristics pipeline, we ran 4 filtering experiments: GitHub stars, tokenizer fertility, comment-to-code ratio and more near-deduplication.

Filtering for GitHub stars hurts performance while comments and near-dedup help!

View quoted post on X

Miltos Allamanis 🇪🇺 @miltos1 Aug 16, 2022

🐶CoRGi #KDD2022

Often, we forget that graph representations are (lossy) projections of a domain into a graph. In our paper, @jyscardioid et al., we present a simple, yet effective, way to incorporate rich node content into the GNN message-passing.

📄 arxiv.org/abs/2110.04866

Miltos Allamanis 🇪🇺 @miltos1 Apr 26, 2022

Excited for our #ICLR2022 Grammformers work:
💡A grammar-based transformer model generates code inserting holes where it is uncertain about the concrete completion.
👉Reduces mistakes (~potential bugs), ie, handles uncertainty, instead of generating 🗑.
📄 arxiv.org/abs/2106.10158

Miltos Allamanis 🇪🇺 @miltos1 Dec 9, 2021

Our recent work with @mmjb86 at #NeurIPS2021
💡Play hide-and-seek: a model learns to hide bugs, another learns to find them
📄 arxiv.org/abs/2105.12787
🕸While code gen LMs are now popular, generating correct-ish code will first require methods that detect bugs in existing code.

arxiv.org Self-Supervised Bug Detection and Repair Machine learning-based program analyses have recently shown the promise of integrating formal and probabilistic reasoning towards aiding software development. However, in the absence of large...

Microsoft Research Verified account @MSFTResearch Dec 8, 2021

Fixing bugs in code can be time-consuming and frustrating for software developers. A promising deep learning model can be taught to detect and fix bugs, without using labeled data, through a “hide and seek” game called BugLab.  #NeurIPS2021 msft.it/6007khWNV
View quoted post on X

Miltos Allamanis 🇪🇺 @miltos1 Dec 1, 2021

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)