Abstract: A cumulative reward of a target system type is predicted by training a prediction model by performing an iteration for each time step. The iteration includes recursively updating a matrix by using the weighted difference between an eligibility trace of a current time step and an eligibility trace of a previous time step, recursively updating a vector by using a reward of a subsequent time step and the eligibility trace of the current time step, and recursively updating an eligibility trace of a subsequent time step by using a feature vector of the subsequent time step. Each feature vector is an encoded representation of a state of a training system of the target system type at a corresponding time step. The matrix and the vector are output as the prediction model for estimating the cumulative reward of a target time step of a target system of the target system type.

Abstract: An n-steps-ahead value of time-series data is estimated by a prediction model configured to output a sum of discounted m-th order differences of adjacent time steps at each time step, wherein an m-th order difference at a corresponding time step is discounted by using a discount factor such that an m-th order difference discount increases as a time step of the m-th order difference increases.

Abstract: A computer-implemented method for balancing loads of a distributed system having a plurality of nodes via a load balancing scheme is presented. The method includes determining an average load of the plurality of nodes once a request is sent to the distributed system, determining a threshold load value based on the determined average load of the plurality of nodes, and randomly selecting a node of the plurality of nodes based on a hash value. The method further includes determining whether the randomly selected node is above or below the threshold load value, and, if the randomly selected node is above the threshold load value, randomly selecting another node, and if the randomly selected node is below the threshold load value, then selecting such node to process the request.

Abstract: A training method is provided. The training method includes clustering, by a processor, a plurality of items that each have an item attribute value, according to the item attribute value. The training method further includes generating, by the processor, for each item, a cluster attribute value corresponding to a cluster associated with the item. The training method also includes training, by the processor, an estimation model for estimating selection behavior of a target with respect to a choice set including two or more items, based on the cluster attribute value associated with each item included in the choice set, by using training data that includes a group of a choice set of items presented to the target and an item selected by the target from among the choice set.

Abstract: A learning apparatus and method for learning a model corresponding to time-series input data, comprising: acquire the time-series input data; supply a plurality of input nodes of the model with a plurality of input values corresponding to input data at one time point in the time-series input data; store values of hidden nodes; compute a conditional probability of each input value at the one time point on a condition that an input data sequence has occurred, based on the input data sequence before the one time point in the time-series input data, on the stored values of hidden nodes, and on weight parameters; and perform a learning process that further increases a conditional probability of input data occurring at the one time point on the condition that the input data sequence has occurred, by adjusting the weight parameters.

Abstract: Methods and systems for game event analysis and control include determining segmentation events in a game that each represent an interruption in game play or a reset scene. An event analysis is performed using the series of events to determine a contribution value for each event in the series of events that represents how much each respective event in the series of events contributed to the point being scored. Events that have a contribution value that exceeds a baseline value for a respective event type are responded to.

Abstract: Regularization of neural networks. Neural networks can be regularized by obtaining an original neural network having a plurality of first-in-first-out (FIFO) queues, each FIFO queue located between a pair of nodes among a plurality of nodes of the original neural network, generating at least one modified neural network, the modified neural network being equivalent to the original neural network with a modified length of at least one FIFO queue, evaluating each neural network among the original neural network and the at least one modified neural network, and determining which neural network among the original neural network and the at least one modified neural network is most accurate, based on the evaluation.

Abstract: A computer-implement method and an apparatus are provided for neural network reinforcement learning. The method includes obtaining, by a processor, an action and observation sequence. The method further includes inputting, by the processor, each of a plurality of time frames of the action and observation sequence sequentially into a plurality of input nodes of a neural network. The method also includes updating, by the processor, a plurality of parameters of the neural network by using the neural network to approximate an action-value function of the action and observation sequence.

Abstract: A method is provided for evaluating a next action of a target object in an environment. The method includes simulating, by a processor device for each of possible actions of the target object in the environment, a next state occurring thereafter to obtain a plurality of simulated next states, based on a pessimistic scenario in which all possible unfavorable actions of other objects occur in the next state in simulation. At least two of the possible unfavorable actions in the next state are unable to simultaneously occur in reality. The method further includes identifying, by the processor device, a safety area for the target object in each of the plurality of simulated next states. The method also includes evaluating, by the processor device, each of the possible actions of the target object, based on the safety area for the target object in each of the plurality of simulated next states.

Abstract: A method detects sports highlights for an event involving players and a play object. The method calculates a covariance matrix characterizing a two-dimensional variation of a multivariate probability distribution presumed for the players relative to a spatiotemporal tracking dataset corresponding to the players. The method calculates an area S occupied by the distribution as a function S(t) based on Eigenvalues calculated from the covariance matrix, derives a differential of the function S(t) and obtains an absolute value of the differential using a function ƒ(t). The method defines a function g(t) relative to a threshold tthreshold. The function g(t) assigns a first or second value responsive to a current time t meeting or exceeding, respectively, the threshold tthreshold. The method outputs as the sports highlight, a highest value from among local peaks extracted from a convolution of the functions ƒ(t) and g(t) and being within a range of the threshold tthreshold.

Abstract: Systems and methods for determining a treatment action include recording batches of data in a replay buffer, each of the batches including a present state, a previous state and a previous action. A value of each action in a set of candidate actions is evaluated at the present state according to a probability that each action achieves a goal of resolving a patient condition or achieves an objective for treating the patient condition by using a value model head corresponding to the goal and the objective. The treatment action is determined from the set of candidate actions according to the value of each action. The treatment action is communicated to a user to treat the patient condition. An error of the value of each action is determined according to whether the previous state achieved by the previous action matches the goal of the objective. Parameters of the value model are updated according to the error.

Abstract: A dynamic time-evolution Boltzmann machine capable of learning is provided. Aspects include acquiring a time-series input data and supplying a plurality of input values of input data of the time-series input data at one time point to a plurality of nodes of the mode. Aspects also include computing, based on an input data sequence before the one time point in the time-series input data and a weight parameter between each of a plurality of input values of input data of the input data sequence and a corresponding one of the plurality of nodes of the model, a conditional probability of the input value at the one time point given that the input data sequence has occurred. Aspects further include adjusting the weight parameter so as to increase a conditional probability of occurrence of the input data at the one time point given that the input data sequence has occurred.

Abstract: Systems and methods for modelling time-series data includes testing a testing model with a plurality of hyper-forgetting rates to select a best performing hyper-forgetting rate. A model optimization is tested using the best performing hyper-forgetting rate with the testing model to test combinations of hyper-parameters to select a best performing combination of hyper-parameters. An error of the model is determined using the model optimization. Model parameters are recursively updated according to the least squares regression by determining a pseudo-inverse of a Hessian of the least squares regression at a current time stamp according to a projection of the time-series data at the current time stamp and the pseudo-inverse of the Hessian at a previous time-stamp to determine an optimum model parameter. A next step behavior of the time-series data is predicted using the optimum model parameter. The next step behavior is stored in a database for access by a user.

Abstract: Methods and systems for game event analysis and control include determining segmentation events in a game that each represent an interruption in game play or a reset scene. An event analysis is performed using the series of events to determine a contribution value for each event in the series of events that represents how much each respective event in the series of events contributed to the point being scored. Events that have a contribution value that exceeds a baseline value for a respective event type are responded to.

Abstract: A computer-implemented method is provided for machine prediction. The method includes forming, by a hardware processor, a Convolutional Dynamic Boltzmann Machine (C-DyBM) by extending a non-convolutional DyBM with a convolutional operation. The method further includes generating, by the hardware processor using the convolution operation of the C-DyBM, a prediction of a future event at time t from a past patch of time-series of observations. The method also includes performing, by the hardware processor, a physical action responsive to the prediction of the future event at time t.

Abstract: Methods and systems for selecting and performing group actions include selecting parameters for an approximated action-value function, which determines a reward value associated with a particular group action taken from a particular state, using a determinant of a parameter matrix for the action-value function. A group action is selected using the approximated action-value function and the selected parameters. Agents are triggered to perform respective tasks in the group action.

Abstract: Fluid motion is simulated by performing a first fluid simulation without reflecting all of a plurality of forces acting on a fluid, to obtain a first velocity of the fluid at a current time step; estimating a velocity residue at the current time step by inputting a calculated velocity from the previous time step into a regression model. The regression model is trained to relate velocity obtained by performing a second fluid simulation reflecting the plurality of forces acting on the fluid to a difference between the velocity obtained by performing the first fluid simulation and the velocity obtained by performing the second fluid simulation; and calculating a velocity of the fluid at the current time step by adding the first velocity at the current time step and the velocity residue at the current time step.

Abstract: A processing apparatus is disclosed for representing cognitively biased selection behavior of a consumer as a learnable model with high prediction accuracy taking into account even feature values of a product and the consumer. The processing apparatus generates a selection model obtained by modeling selection behavior of a selection entity that selects at least one choice out of presented input choices. The processing apparatus includes an acquiring unit to acquire training data including a plurality of input feature vectors that indicate features of a plurality of the choices presented to the selection entity and an output feature vector that indicates a feature of an output choice. The processing apparatus further includes an input combining unit to combine the plurality of input vectors to generate an input combined vector, and a learning processing unit to learn a selection model on the basis of the input combined vector and the output vector.

Abstract: A method is provided for reducing a computational cost of deep reinforcement learning using an input image to provide a filtered output image composed of pixels. The method includes generating a moving gate in which the pixels of the filtered output image to be masked are assigned a first gate value and the pixels of the filtered output image to be passed through are assigned a second gate value. The method further includes applying the input image and the moving gate to a GCNN to provide the filtered output image such that only the pixels of the input image used to compute the pixels assigned the second gate value are processed by the GCNN while bypassing the pixels of the input image useable to compute the pixels assigned the first gate to reduce an overall processing time of the input image in order to provide the filtered output image.

Abstract: Provided are a computer program product, a learning apparatus and a learning method. The method includes calculating a first propagation value that is propagated from a propagation source node to a propagation destination node in a neural network including nodes, based on node values of the propagation source node at time points and a weight corresponding to passage of time points based on a first attenuation coefficient. The method also includes updating the first attenuation coefficient by using a first update parameter, that is based on a first propagation value, and an error of the node value of the propagation destination node.