Generation of computer code using look-ahead tree research controlled by a neural network
DE202026104733U1Undetermined Publication Date: 2026-11-26GDM HOLDING LLC
0 Cites 0 Cited by
Patent Information
- Application Number
- DE202026104733
- Authority / Receiving Office
- DE · DE
- Patent Type
- Utility models
- Current Assignee / Owner
- Priority Date
- 2025-08-06
- Filing Date
- 2026-08-06
- Publication Date
- 2026-11-26
- Estimated Expiration
- 2036-08-31
Abstract
A system comprising one or more computing devices and one or more storage devices that store instructions which, when executed by the one or more computing devices, cause the one or more computing devices to perform operations that include: obtaining an initial sequence of computer code to perform a computer programming task;Generating a final sequence of computer code to perform the computer programming task by performing a look-ahead tree search controlled by a neural network to traverse a tree representing a search space of computer code to perform the computer programming task, wherein the look-ahead tree search comprises a plurality of search iterations and wherein performing the look-ahead tree search at each search iteration in the plurality of search iterations comprises: traversing the tree, starting from a root node of the tree representing the initial sequence of computer code, until reaching a leaf node in the tree;Processing an input comprising a current sequence of computer code represented by the leaf node, using an encoder neural network in the neural network to generate a latent representation of the current sequence of computer code represented by the leaf node; processing an input comprising the latent representation of the current sequence of computer code represented by the leaf node, using a policy head in the neural network to generate a policy output comprising a plurality of instructions, each instruction defining a respective direction for modifying the current sequence of computer code represented by the leaf node;and updating the tree by adding to the tree, for each of a subset of the plurality of instructions, (i) a respective child node of the leaf node, representing a respective modified sequence of computer code that can be generated by modifying the current sequence of computer code represented by the leaf node in accordance with the respective direction defined by the instruction, and (ii) a respective edge between the leaf node and the respective child node of the leaf node.
Need to check novelty before this filing date? Find Prior Art
