Abstract: A method develops a programming language model for machine learning tasks. The method includes adjusting a token list to include a language token used by a tokenizer for a pretrained language model. The pretrained language model includes a set of layers. The set of layers includes a set of initial layers, an embedding layer, and an output layer. The method further includes performing an output layer modification of the output layer to replace the output vector with the embedding vector. The method further includes freezing the set of initial layers to generate a set of frozen layers of the pretrained language model that do not update during training. The method further includes training the pretrained language model using the language token, the output layer modification, and the set of frozen layers to form a fine-tuned model from the pretrained language model.

Abstract: A method includes disassembling a reference binary of a library to generate a control flow graph of the referenced binary, normalizing the control flow graph to generate a normalized graph, traversing the normalized graph to generate execution traces from the normalized graph, and generating library vector embeddings. Generating library vector embeddings includes, for each execution trace of at least a subset of the execution traces, processing the execution trace by a vector embedding model to generate a library vector embedding of the execution trace. The method further includes relating, in storage, a library identifier of the library to the plurality of library vector embeddings as a fingerprint of the library.

Abstract: A method of code vulnerability detection includes obtaining a set of patches from at least one code repository for a vulnerability record of a vulnerability of a software component, applying the set of patches to obtain a modified set of code files modified by the set of patches, and inferring a vulnerable code section set according to the modified set of code files. The vulnerable code section set is a subset of the modified set of code files. The method further includes obtaining a vulnerable version list of a plurality of versions of the software component comprising the vulnerability, validating a version in the vulnerable version list as having the vulnerable code section set, and processing a target software comprising the version using the vulnerable code section set.

Abstract: A system may include a computer processor and a repository configured to store a first code fragment including language features represented in a first programming language, and a second code fragment including language features represented in a second programming language. The system may further include a universal code fragment classifier, executing on the computer processor and configured to generate a first universal abstract syntax tree for the first code fragment and a second universal abstract syntax tree for the second code fragment, generate, using a graph embedding model, first vectors for the first universal abstract syntax tree and second vectors for the second universal abstract syntax tree, and classify, by executing an abstract syntax tree classifier on the first vectors and the second vectors, the first code fragment as a first code category and the second code fragment as a second code category.

Abstract: A method may include extracting, from a function included in code, sub-tokens and program analysis features, generating sub-token vectors from the sub-tokens and a program analysis vector from the program analysis features, combining, by a machine learning model, the sub-token vectors to obtain a combined sub-token vector, combining the combined sub-token vector and the program analysis vector to obtain a function vector, and classifying, using the function vector, the function as a function category.

Abstract: A system may include a computer processor and a repository configured to store a first code fragment including language features represented in a first programming language, and a second code fragment including language features represented in a second programming language. The system may further include a universal code fragment classifier, executing on the computer processor and configured to generate a first universal abstract syntax tree for the first code fragment and a second universal abstract syntax tree for the second code fragment, generate, using a graph embedding model, first vectors for the first universal abstract syntax tree and second vectors for the second universal abstract syntax tree, and classify, by executing an abstract syntax tree classifier on the first vectors and the second vectors, the first code fragment as a first code category and the second code fragment as a second code category.

Abstract: A method may including obtaining, for an application, an application dependency including called components, and obtaining, using the called components, a component compatibility graph including a set of nodes and a set of edges each connecting a pair of nodes in the set of nodes. The pair of nodes may include a calling node and a called node. Each node may correspond to a component. The method may further include generating, from the component compatibility graph, a set of constraints including a set of edge variables corresponding to the set of edges, selecting, using the set of constraints, an edge subset of the set of edges, and recommending, for the application, an upgrade solution including installing a called component corresponding to a called node connected by an edge in the edge subset.

Abstract: A method may including obtaining, for an application, an application dependency including called components, and obtaining, using the called components, a component compatibility graph including a set of nodes and a set of edges each connecting a pair of nodes in the set of nodes. The pair of nodes may include a calling node and a called node. Each node may correspond to a component. The method may further include generating, from the component compatibility graph, a set of constraints including a set of edge variables corresponding to the set of edges, selecting, using the set of constraints, an edge subset of the set of edges, and recommending, for the application, an upgrade solution including installing a called component corresponding to a called node connected by an edge in the edge subset.

Abstract: A system may include a computer processor and a repository configured to store a first code fragment including language features represented in a first programming language, and a second code fragment including language features represented in a second programming language. The system may further include a universal code fragment classifier, executing on the computer processor and configured to generate a first universal abstract syntax tree for the first code fragment and a second universal abstract syntax tree for the second code fragment, generate, using a graph embedding model, first vectors for the first universal abstract syntax tree and second vectors for the second universal abstract syntax tree, and classify, by executing an abstract syntax tree classifier on the first vectors and the second vectors, the first code fragment as a first code category and the second code fragment as a second code category.

Abstract: A method may include extracting, from a function included in code, sub-tokens and program analysis features, generating sub-token vectors from the sub-tokens and a program analysis vector from the program analysis features, combining, by a machine learning model, the sub-token vectors to obtain a combined sub-token vector, combining the combined sub-token vector and the program analysis vector to obtain a function vector, and classifying, using the function vector, the function as a function category.

Abstract: A system may include a computer processor and a repository configured to store a first code fragment including language features represented in a first programming language, and a second code fragment including language features represented in a second programming language. The system may further include a universal code fragment classifier, executing on the computer processor and configured to generate a first universal abstract syntax tree for the first code fragment and a second universal abstract syntax tree for the second code fragment, generate, using a graph embedding model, first vectors for the first universal abstract syntax tree and second vectors for the second universal abstract syntax tree, and classify, by executing an abstract syntax tree classifier on the first vectors and the second vectors, the first code fragment as a first code category and the second code fragment as a second code category.