Transformer with Tree-order Encoding for Neural Program Generation

Research output: Contributions to collected editions/worksArticle in conference proceedingsResearch

Standard

Transformer with Tree-order Encoding for Neural Program Generation. / Thellmann, Klaudia-Doris; Stadler, Bernhard; Usbeck, Ricardo et al.
Conference XXX. 2022.

Research output: Contributions to collected editions/worksArticle in conference proceedingsResearch

Harvard

APA

Vancouver

Thellmann KD, Stadler B, Usbeck R, Lehmann J. Transformer with Tree-order Encoding for Neural Program Generation. In Conference XXX. 2022 doi: 10.48550/arXiv.2206.13354

Bibtex

@inbook{56b3c87266394178bbbbf7097e04d0af,
title = "Transformer with Tree-order Encoding for Neural Program Generation",
abstract = " While a considerable amount of semantic parsing approaches have employed RNN architectures for code generation tasks, there have been only few attempts to investigate the applicability of Transformers for this task. Including hierarchical information of the underlying programming language syntax has proven to be effective for code generation. Since the positional encoding of the Transformer can only represent positions in a flat sequence, we have extended the encoding scheme to allow the attention mechanism to also attend over hierarchical positions in the input. Furthermore, we have realized a decoder based on a restrictive grammar graph model to improve the generation accuracy and ensure the well-formedness of the generated code. While we did not surpass the state of the art, our findings suggest that employing a tree-based positional encoding in combination with a shared natural-language subword vocabulary improves generation performance over sequential positional encodings. ",
keywords = "cs.CL, cs.AI, 68T07, 68T50, I.2.7, Informatics",
author = "Klaudia-Doris Thellmann and Bernhard Stadler and Ricardo Usbeck and Jens Lehmann",
note = "This paper was authored in late 2020 and early 2021 for the most part",
year = "2022",
month = may,
day = "30",
doi = "10.48550/arXiv.2206.13354",
language = "English",
booktitle = "Conference XXX",

}

RIS

TY - CHAP

T1 - Transformer with Tree-order Encoding for Neural Program Generation

AU - Thellmann, Klaudia-Doris

AU - Stadler, Bernhard

AU - Usbeck, Ricardo

AU - Lehmann, Jens

N1 - This paper was authored in late 2020 and early 2021 for the most part

PY - 2022/5/30

Y1 - 2022/5/30

N2 - While a considerable amount of semantic parsing approaches have employed RNN architectures for code generation tasks, there have been only few attempts to investigate the applicability of Transformers for this task. Including hierarchical information of the underlying programming language syntax has proven to be effective for code generation. Since the positional encoding of the Transformer can only represent positions in a flat sequence, we have extended the encoding scheme to allow the attention mechanism to also attend over hierarchical positions in the input. Furthermore, we have realized a decoder based on a restrictive grammar graph model to improve the generation accuracy and ensure the well-formedness of the generated code. While we did not surpass the state of the art, our findings suggest that employing a tree-based positional encoding in combination with a shared natural-language subword vocabulary improves generation performance over sequential positional encodings.

AB - While a considerable amount of semantic parsing approaches have employed RNN architectures for code generation tasks, there have been only few attempts to investigate the applicability of Transformers for this task. Including hierarchical information of the underlying programming language syntax has proven to be effective for code generation. Since the positional encoding of the Transformer can only represent positions in a flat sequence, we have extended the encoding scheme to allow the attention mechanism to also attend over hierarchical positions in the input. Furthermore, we have realized a decoder based on a restrictive grammar graph model to improve the generation accuracy and ensure the well-formedness of the generated code. While we did not surpass the state of the art, our findings suggest that employing a tree-based positional encoding in combination with a shared natural-language subword vocabulary improves generation performance over sequential positional encodings.

KW - cs.CL

KW - cs.AI

KW - 68T07, 68T50

KW - I.2.7

KW - Informatics

U2 - 10.48550/arXiv.2206.13354

DO - 10.48550/arXiv.2206.13354

M3 - Article in conference proceedings

BT - Conference XXX

ER -

Recently viewed

Publications

  1. Learning Rotation Sensitive Neural Network for Deformed Objects' Detection in Fisheye Images
  2. Selecting and Adapting Methods for Analysis and Design in Value-Sensitive Digital Social Innovation Projects: Toward Design Principles
  3. Invariant subspaces for grasping internal forces and non-interacting force-motion control in robotic manipulation
  4. Classical PI Controllers with Anti-Windup Techniques Applied on Level Systems
  5. Lagged Multidimensional Recurrence Quantification Analysis for Determining Leader–Follower Relationships Within Multidimensional Time Series
  6. Effective informational entropy reduction in multi-robot systems based on real-time TVS
  7. Positioning Improvement for a Laser Scanning System using cSORPD control
  8. Improved sensorimotor control is not connected with improved proprioception
  9. Machine Learning and Knowledge Discovery in Databases
  10. Competing Vegetation Structure Indices for Estimating Spatial Constrains in Carabid Abundance Patterns in Chinese Grasslands Reveal Complex Scale and Habitat Patterns
  11. Advances in Dynamics, Optimization and Computation
  12. Does thinking-aloud affect learning, visual information processing and cognitive load when learning with seductive details as expected from self-regulation perspective?
  13. Cognitive load and instructionally supported learning with provided and learner-generated visualizations
  14. Resource extraction technologies - is a more responsible path of development possible?
  15. Using augmented video to test in-car user experiences of context analog HUDs
  16. Robust Control of Mobile Transportation Object with 3D Technical Vision System
  17. How Much Home Office is Ideal? A Multi-Perspective Algorithm
  18. Efficient Order Picking Methods in Robotic Mobile Fulfillment Systems
  19. Guided discovery learning with computer-based simulation games
  20. Probabilistic approach to modelling of recession curves
  21. Eighth Workshop on Mining and Learning with Graphs
  22. Mostly harmless econometrics? Statistical paradigms in the ‘top five’ from 2000 to 2018
  23. Learning and Re-learning from net- based cooperative learning discourses
  24. Using Heider’s Epistemology of Thing and Medium for Unpacking the Conception of Documents: Gantt Charts and Boundary Objects
  25. Topic Embeddings – A New Approach to Classify Very Short Documents Based on Predefined Topics
  26. Transfer operator-based extraction of coherent features on surfaces
  27. Optimising business performance with standard software systems
  28. Reality-Based Tasks with Complex-Situations
  29. On the Difficulty of Forgetting
  30. An experience-based learning framework
  31. Soft Skills for Hard Constraints
  32. Should learners use their hands for learning? Results from an eye-tracking study