English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Program synthesis or code generation aims to generate a program thatsatisfies a problem specification. Recent approaches using large-scalepretrained language models (LMs) have shown promising results, yet they havesome critical limitations. In particular, they often follow a standardsupervised fine-tuning procedure to train a code generation model only from thepairs of natural-language problem descriptions and ground-truth programs. Suchparadigm largely ignores some important but potentially useful signals in theproblem specification such as unit tests, which thus often results in poorperformance when solving complex unseen coding tasks. To address thelimitations, we propose "CodeRL", a new framework for program synthesis tasksthrough pretrained LMs and deep reinforcement learning (RL). Specifically,during training, we treat the code-generating LM as an actor network, andintroduce a critic network that is trained to predict the functionalcorrectness of generated programs and provide dense feedback signals to theactor. During inference, we introduce a new generation procedure with acritical sampling strategy that allows a model to automatically regenerateprograms based on feedback from example unit tests and critic scores. For themodel backbones, we extended the encoder-decoder architecture of CodeT5 withenhanced learning objectives, larger model sizes, and better pretraining data.Our method not only achieves new SOTA results on the challenging APPSbenchmark, but also shows strong zero-shot transfer capability with new SOTAresults on the simpler MBPP benchmark.

CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning