Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating a plan for an input task. The method involves receiving a query for the input task and obtaining candidate task-plan examples. An initial plan, comprising an initial action sequence, is obtained for the input task. For each candidate example, an action sequence similarity score is computed, measuring similarity between the initial action sequence and the action sequence in the candidate example. A set of task-plan examples is selected from the candidates using these similarity scores. Finally, a generative machine learning model processes the query and prompt inputs generated from the selected set of examples to generate an output plan. This approach improves plan generation by selecting relevant examples based on procedural similarity rather than superficial task description similarity, enhancing the quality and accuracy of the output plan.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a neural network. One of the methods includes obtaining data specifying a trained neural network that includes a plurality of layers that include a particular layer; generating an adapted neural network, comprising generating, for the particular layer, an approximation of an adapter parameter matrix that includes fewer parameters than the adapter parameter matrix; and training the adapted neural network on a machine learning task, wherein the adapting comprises learning fine-tuned values of parameters of the approximation using training data while holding the trained values in the base parameter matrix fixed.

Abstract: Methods, systems, and apparatus for adaptively generating test stimuli for testing a hardware design for an integrated circuit. In one aspect, a system comprises one or more computers configured to obtain graph data representing a coverage dependency graph associated with a hardware design for an integrated circuit. At each of a plurality of simulation cycles, the one or more computers obtain a set of coverage statistics as of the simulation cycle and update respective distribution constraints for one or more random variables in a set of random variables using the coverage dependency graph and the coverage statistics. After the updating, the one or more computers generate one or more test stimuli by, for each test stimulus, sampling a respective value for each of the random variables based on the respective distribution constraints. The one or more computers simulate a performance of the integrated circuit for each of the test stimuli.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing a machine learning task on a network input to generate a network output. One of the methods includes: obtaining data specifying an initial neural network configured to perform a machine learning task; a representativeness measure for each of a plurality of filters; determining a central tendency measure for the plurality of filters based on processing a batch of network inputs using the initial neural network; determining a cumulative importance score for each of the plurality of filters; selecting a proper subset of the plurality of filters; and generating a pruned neural network configured to perform the machine learning task.