Abstract: Methods, systems, and apparatus for obtaining input features representative of a region of space, processing an input comprising the input features through the ML model to generate a prediction describing predicted features of the region of space, obtaining result features describing the region of space, determining a value of at least one evaluation metric that relates the predicted features and the result features, that at least one evaluation metric including one of a distance score, a pyramiding density error, and min-max intersection over union (IOU) score, and training the ML model responsive to the at least one evaluation metric. Other implementations of this aspect include corresponding systems, apparatus, and computer programs, configured to perform the actions of the methods, encoded on computer storage devices.

Abstract: Training and/or utilization of machine learning model(s) (e.g., neural network model(s)) in automatically generating test case(s) for source code. Techniques disclosed herein can be utilized in generating test case(s) for unit test testing (or other white-box testing) and/or for functional testing (or other black-box testing).

Abstract: Techniques are described herein for translating source code in one programming language to source code in another programming language using machine learning. A method includes: receiving first source code in a first higher-level programming language; processing the first source code, or an intermediate representation thereof, using a sequence-to-sequence neural network model to generate a sequence of outputs, each including a probability distribution; generating second source code in a second higher-level programming language by, for each output in the sequence of outputs: determining a highest probability in the probability distribution associated with the output; in response to the highest probability exceeding a first threshold, generating a predicted portion of the second source code based on a token that corresponds to the highest probability; and in response to the highest probability not exceeding the first threshold, generating a placeholder; and outputting the second source code.

Abstract: Disclosed implementations relate to using mutual constraint satisfaction to sample from different stochastic processes and identify coherent inferences across domains. In some implementations, a first domain representation of a semantic concept may be used to conditionally sample a first set of candidate second domain representations of the semantic concept from a first stochastic process. Based on second domain representation(s) of the first set, candidate third domain representations of the semantic concept may be conditionally sampled from a second stochastic process. Based on candidate third domain representation(s), a second set of candidate second domain representations of the semantic concept may be conditionally sampled from a third stochastic process. Pairs of candidate second domain representations sampled across the first and second sets may be evaluated. Based on the evaluation, second domain representation(s) of the semantic concept are selected, e.g., as input for a downstream computer process.

Abstract: Aspects of the disclosure relate to identifying potential areas of innovations utilizing machine learning models such as LLMs. As an example, a plurality of application areas and a plurality of sources may be identified. A model may be used to score pairs of each one of the plurality of application areas with respect to each one the plurality of sources. A matrix of the scores may be generated and provided for display to a user.

Abstract: Implementations are described herein for aggregating information responsive to a query from multiple different data feed services using machine learning. In various implementations, NLP may be performed on a natural language input comprising a query for information to generate a data feed-agnostic aggregator embedding (FAAE). A plurality of data feed services may be selected, each having its own data feed service action space that includes actions that are performable to access data via the data feed service. The FAAE may be processed based on domain-specific machine learning models corresponding to the selected data feed services. Each domain-specific machine learning model may translate between a respective data feed service action space and a data feed-agnostic semantic embedding space. Using these models, action(s) may be selected from the data feed service action spaces and performed to aggregate, from the plurality of data feed services, data that is responsive to the query.

Abstract: Disclosed implementations relate to automating semantically-similar computing tasks across multiple contexts. In various implementations, an initial natural language input and a first plurality of actions performed using a first computer application may be used to generate a first task embedding and a first action embedding in action embedding space. An association between the first task embedding and first action embedding may be stored. Later, subsequent natural language input may be used to generate a second task embedding that is then matched to the first task embedding. Based on the stored association, the first action embedding may be identified and processed using a selected domain model to select actions to be performed using a second computer application. The selected domain model may be trained to translate between an action space of the second computer application and the action embedding space.

Abstract: The present invention relates to a process for producing hydrogen peroxide by the AO process, comprising the two alternating steps of: hydrogenation of a working solution in the presence of a catalyst, said working solution containing at least one quinone dissolved in at least one organic solvent, in order to obtain at least one corresponding hydroquinone; and oxidation of said at least one hydroquinone; the solvent corresponding to formula (I) below: in which n is an integer greater than or equal to 8.

Abstract: Disclosed herein are systems and methods for objectively characterizing machine-learning models including receiving first training data formatted to be used in the training of a machine-learning model; receiving one or more challenge queries formatted to be run on the machine-learning model; generating, for the first training data, a plurality of associated training vectors that embed at least some of the first training data into a vector space; generating, for each of the one or more challenge queries, a plurality of associated challenge vectors that embed at least some of the challenge queries into the vector space; and determining, for each challenge query, a corresponding quality metric for the machine-learning model by determining a neighborhood density for each of the challenge queries in the vector space.

Abstract: This specification is generally directed to techniques for creating reduced-dimensionality embeddings (e.g., embedding layers of neural networks) with dimensions that are interpretable by and/or are meaningful to humans. In various implementations, a datum may sampled from a document. A dimensionality reduction process may be performed based on the sampled datum to generate a semantically-interpretable embedding having a number of individually-interpretable dimensions. The dimensionality reduction process may include: analyzing the sampled datum according to a number of distinct semantic queries to determine respective numeric solutions. The number of distinct semantic queries may correspond to the number of individually-interpretable dimensions. Each numeric solution may offer an inconclusive clue about the sampled datum. The dimensionality reduction process may also include populating the dimensions of the semantically-interpretable embedding with respective numeric solutions.

Abstract: Disclosed implementations relate to automating semantically-similar computing tasks across multiple contexts. In various implementations, an initial natural language input and a first plurality of actions performed using a first computer application may be used to generate a first task embedding and a first action embedding in action embedding space. An association between the first task embedding and first action embedding may be stored. Later, subsequent natural language input may be used to generate a second task embedding that is then matched to the first task embedding. Based on the stored association, the first action embedding may be identified and processed using a selected domain model to select actions to be performed using a second computer application. The selected domain model may be trained to translate between an action space of the second computer application and the action embedding space.

Abstract: This specification is generally directed to techniques for robust natural language (NL) based control of computer applications. In many implementations, the NL control is at least selectively interactive in that the user feedback input is solicited, and received, in resolving action(s), resolving action set(s), generating domain specific knowledge, and/or in providing feedback on implemented action set(s). The user feedback input can be utilized in further training of machine learning model(s) utilized in the NL based control of the computer applications.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating actions for a supply chain network. One of the methods includes receiving a request to generate an action in a supply chain network for a particular product based on current state information; providing a request to an action model to generate a respective probability distribution for one or more actions for one or more products; receiving, from the action model, the respective probability distributions for the one or more products; determining, for each product, a binned action from the respective probability distribution; providing a request to a sequence model to generate a respective correction for the one or more binned actions; and receiving, from the sequence model, the respective correction for the respective binned action.

Abstract: An inverse design system combines a large language model (LLM) with a task-specific optimizer, which includes a search function, a forward model, and a comparator. The LLM adjusts parameters of the optimizer's components in response to a design scenario. Then the optimizer processes the design scenario to produce design candidates. Optionally, the LLM learns from the design candidates in an iterative process. A stochastic predictive modeling system combines an LLM with input distributions and a forward model. The LLM adjusts one or more of the input distributions and/or the forward model in response to a forecast scenario. Then the forward model processes a sampling of the input distributions to produce a forward distribution. Optionally, the LLM informs the sampling process. Optionally, the LLM learns from the forward distribution.

Abstract: This specification is generally directed to techniques for generating interfacing source code between computing components based on natural language input. In various implementations, a natural language input that requests generation of interfacing source code to logically couple a first computing component with a second computing component may be processed to generate an interface request semantic embedding. The interface request semantic embedding may be processed based on one or more domain models associated with the first and second computing components to generate a pool(s) of candidate code snippets for logically coupling with first and second computing components. A plurality of candidate instances of interfacing source code may be generated between the first and second computing components. Each candidate software interface may include a different permutation of candidate code snippets from the pool(s) of candidate code snippets.

Abstract: Disclosed implementations relate to adding “bottleneck” models to machine learning pipelines that already apply domain models to translate and/or transfer representations of high-level semantic concepts between domains. In various implementations, an initial representation in a first domain of a transition from an initial state of an environment to a goal state of the environment may be processed based on a pre-trained first domain encoder to generate a first embedding that semantically represents the transition. The first embedding may be processed based on one or more bottleneck models to generate a second embedding with fewer dimensions than the first embedding. In various implementations, the second embedding may be processed in various ways to train one or more of the bottleneck model(s) based on various different auxiliary loss functions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing tasks. One of the methods includes obtaining a prompt, obtaining a set of documents, generating an input, providing the input to a plurality of language models, generating a distribution from intermediate answers from the language models; and generating an answer to the prompt by performing a probabilistic inference over the distribution.

Abstract: Disclosed implementations relate to preserving individuals' semantic privacy while facilitating automation of tasks across a population of individuals. In various implementations, data indicative of an observed set of interactions between a user and a computing device may be recorded and used to simulate multiple different synthetic sets of interactions between the user and the computing device. Each synthetic set may include a variation of the observed set of interactions at a different level of abstraction. User feedback may be obtained about each of the multiple different sets. Based on the user feedback, one of the multiple different synthetic sets of interactions may be selected and used to train a machine learning model.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating synthetic training data representing network disruptions. One of the methods includes obtaining data representing one or more first travel time distributions between at the at least two entities in the supply chain network. Synthetic network disruption data is generated including sampling from one or more second travel time distributions corresponding respectively to one or more simulated network disruptions. A second dataset having the synthetic network disruption data is generated, and a network policy agent is trained using the second dataset.

Abstract: Implementations are described herein for aggregating information responsive to a query from multiple different data feed services using machine learning. In various implementations, NLP may be performed on a natural language input comprising a query for information to generate a data feed-agnostic aggregator embedding (FAAE). A plurality of data feed services may be selected, each having its own data feed service action space that includes actions that are performable to access data via the data feed service. The FAAE may be processed based on domain-specific machine learning models corresponding to the selected data feed services. Each domain-specific machine learning model may translate between a respective data feed service action space and a data feed-agnostic semantic embedding space. Using these models, action(s) may be selected from the data feed service action spaces and performed to aggregate, from the plurality of data feed services, data that is responsive to the query.