Abstract: Techniques for using machine learning (ML) for image processing are disclosed. First encoded image data, generated by encoding a first one or more digital images using an encoder, is received. A first reconstructed one or more digital images are generated by decoding the encoded image data using a decoder corresponding to the encoder. A second reconstructed one or more digital images are generated by transforming the first reconstructed one or more digital images using a super-resolution ML model. The second reconstructed one or more digital images has a higher image resolution compared with the first reconstructed one or more digital images, and the super-resolution ML model is trained based on an image resolution corresponding to at least one of the second reconstructed one or more digital images.

Abstract: A processor training a reinforcement learning model can include receiving a first dataset representing an observable state in reinforcement learning to train a machine to perform an action. The processor receives a second dataset. Using the second dataset, the processor trains a machine learning classifier to make a prediction about an entity related to the action. The processor extracts an embedding from the trained machine learning classifier, and augments the observable state with the embedding to create an augmented state. Based on the augmented state, the processor trains a reinforcement learning model to learn a policy for performing the action, the policy including a mapping from state space to action space.

Abstract: An exemplary method includes solving on a computing system an optimal power flow formulation for a plurality of generators in a power system. The solving includes computing using multi-threaded parallelism a plurality of constraints for the formulation, computing using multi-threaded parallelism a plurality of Jacobian functions of the constraints, and computing using multi-threaded parallelism a Hessian of Lagrangian functions. The method further includes outputting results of the solving, wherein the results comprise values of generation levels for the plurality of generators. Apparatus and program products are also disclosed.

Abstract: A computer-implemented method for generating an abstract meaning representation (“AMR”) of a sentence, comprising receiving, by a computing device, an input sentence and parsing the input sentence into one or more syntactic and/or semantic graphs. An input graph including a node set and an edge set is formed from the one or more syntactic and/or semantic graphs. Node representations are generated by natural language processing. The input graph is provided to a first neural network to provide an output graph having learned node representations aligned with the node representations in the input graph. The method further includes predicting via a second neural network, node label and predicting, via a third neural network, edge labels in the output graph. The AMR is generated based on the predicted node labels and predicted edge labels. A system and a non-transitory computer readable storage medium are also disclosed.

Abstract: An input transformation function that transforms input data for a second machine learning system is learned using a first machine learning system, the learning being based on minimizing a summation of a task loss and a post-activation density loss. The input data is transformed using the learned input transformation function to alter the post-activation density to reduce an amount of energy consumed for an inferencing task and the inferencing task is carried out on the transformed input data using the second machine learning system.

Abstract: A computer-implemented method is provided of using a machine learning model for disentanglement of prosody in spoken natural language. The method includes encoding, by a computing device, the spoken natural language to produce content code. The method further includes resampling, by the computing device without text transcriptions, the content code to obscure the prosody by applying an unsupervised technique to the machine learning model to generate prosody-obscured content code. The method additionally includes decoding, by the computing device, the prosody-obscured content code to synthesize speech indirectly based upon the content code.

Abstract: Techniques regarding action detection based on motion in receptive fields of a neural network model are provided. For example, one or more embodiments described herein can comprise a system, which can comprise a memory that can store computer executable components. The system can also comprise a processor, operably coupled to the memory, and that can execute the computer executable components stored in the memory. The computer executable components can comprise a motion component that can extract a motion vector from a plurality of adaptive receptive fields in a deformable convolution layer of a neural network model. The computer executable components can also comprise an action detection component that can generate a spatio-temporal feature by concatenating the motion vector with a spatial feature extracted from the deformable convolution layer.

Abstract: One or more metrics associated with performance of a serverless function chain are received. The one or more metrics are used by a reinforcement learning (RL) model to tune the serverless function chain.

Abstract: One or more machine learning models for a network of tensor time series can be provided. Co-evolving time series having multiple modes can be received. A tensor graph convolutional network can be trained, using the co-evolving time series and adjacency matrices associated with the multiple modes in the co-evolving time series, to generate node embeddings associated with a snapshot of the co-evolving time series at time t. A tensor recurrent neural network can be trained to generate temporal dynamics associated with the co-evolving time series based on the generated node embeddings. A neural network model can be trained to forecast a prediction for the co-evolving time series based on the generated node embeddings and the generated temporal dynamics. The tensor graph convolutional network, the tensor recurrent neural network and the neural network model can be trained jointly.

Abstract: Time series data can be received. A machine learning model can be trained using the time series data. A contaminating process can be estimated based on the time series data, the contaminating process including outliers associated with the time series data. A parameter associated with the contaminating process can be determined. Based on the trained machine learning model and the parameter associated with the contaminating process, a single-valued metric can be determined, which represents an impact of the contaminating process on the machine learning model's future prediction. A plurality of different outlier detecting machine learning models can be used to estimate the contaminating process and the single-valued metric can be determined for each of the plurality of different outlier detecting machine learning models. The plurality of different outlier detecting machine learning models can be ranked according to the associated single-valued metric.

Abstract: A method includes detecting, by a webhook controller, an inference serverless function invocation. The method further includes determining that the inference serverless function can be optimized. The method further includes generating an optimized version of the inference serverless function using a graph compiler, in response to the determining. The method further includes replacing, by a processing device of the webhook controller, a storage volume in an init container of the inference serverless function with a new storage volume comprising the optimized version of the inference serverless function.

Abstract: One or more computer processors generate a plurality of adversarial perturbations associated with a model, wherein the plurality of adversarial perturbations comprises a universal perturbation and one or more per-sample perturbations. The one or more computer processors identify a plurality of neuron activations associated with the model and the plurality of generated adversarial perturbations. The one or more computer processors maximize the identified plurality of neuron activations. The one or more computer processors determine the model is a Trojan model by leveraging one or more similarities associated with the maximized neuron activations and the generated adversarial perturbations.

Abstract: One or more metrics associated with performance of a serverless function chain are received. The one or more metrics are analyzed by a reinforcement learning (RL) model to determine a tuning factor in view of a goal provided to the RL model. The tuning factor is transmitted to a serverless function controller, wherein the serverless function controller is to utilize the tuning factor to tune the serverless function chain.

Abstract: A computer-implemented method for generating an abstract meaning representation (“AMR”) of a sentence, comprising receiving, by a computing device, an input sentence and parsing the input sentence into one or more syntactic and/or semantic graphs. An input graph including a node set and an edge set is formed from the one or more syntactic and/or semantic graphs. Node representations are generated by natural language processing. The input graph is provided to a first neural network to provide an output graph having learned node representations aligned with the node representations in the input graph. The method further includes predicting via a second neural network, node label and predicting, via a third neural network, edge labels in the output graph. The AMR is generated based on the predicted node labels and predicted edge labels. A system and a non-transitory computer readable storage medium are also disclosed.

Abstract: An exemplary method includes solving on a computing system an optimal power flow formulation for a plurality of generators in a power system. The solving includes computing using multi-threaded parallelism a plurality of constraints for the formulation, computing using multi-threaded parallelism a plurality of Jacobian functions of the constraints, and computing using multi-threaded parallelism a Hessian of Lagrangian functions. The method further includes outputting results of the solving, wherein the results comprise values of generation levels for the plurality of generators. Apparatus and program products are also disclosed.

Abstract: Embodiments for providing critical path analysis of active trace files in a cloud computing environment. A critical path may be identified using a trace of time spans and activities of a plurality of applications, wherein the critical path is a set of activities having time spans free of overlap with other activities.

Abstract: A computer-implemented method is provided for transferring a target text style using Reinforcement Learning (RL). The method includes pre-determining, by a Long Short-Term Memory (LSTM) Neural Network (NN), the target text style of a target-style natural language sentence. The method further includes transforming, by a hardware processor using the LSTM NN, a source-style natural language sentence into the target-style natural language sentence that maintains the target text style of the target-style natural language sentence. The method also includes calculating an accuracy rating of a transformation of the source-style natural language sentence into the target-style natural language sentence based upon rewards relating to at least the target text style of the source-style natural language sentence.

Abstract: A processor training a reinforcement learning model can include receiving a first dataset representing an observable state in reinforcement learning to train a machine to perform an action. The processor receives a second dataset. Using the second dataset, the processor trains a machine learning classifier to make a prediction about an entity related to the action. The processor extracts an embedding from the trained machine learning classifier, and augments the observable state with the embedding to create an augmented state. Based on the augmented state, the processor trains a reinforcement learning model to learn a policy for performing the action, the policy including a mapping from state space to action space.

Abstract: A method includes determining a first upload start timestamp and a first upload end timestamp of a video file from a client device to an object storage device. The method further includes monitoring an upload of the video file from the object storage device to a server-less framework to determine a second upload start timestamp and a second upload end timestamp of the video file. The method further includes determining, by a processing device, a latency of the upload in view of the first upload timestamp, the second upload timestamp, the first upload end timestamp, and the second upload end timestamp. The method further includes providing a latency adjustment instruction to the client device in view of the latency.

Abstract: One or more computer processors generate a plurality of adversarial perturbations associated with a model, wherein the plurality of adversarial perturbations comprises a universal perturbation and one or more per-sample perturbations. The one or more computer processors identify a plurality of neuron activations associated with the model and the plurality of generated adversarial perturbations. The one or more computer processors maximize the identified plurality of neuron activations. The one or more computer processors determine the model is a Trojan model by leveraging one or more similarities associated with the maximized neuron activations and the generated adversarial perturbations.