Abstract: Embodiments are provided for providing enhanced reasoning in a computing system by a processor. All first-order logic formulas may be converted into real-valued logic formulas. A probabilistic inference is executed using the real-valued logic formulas and one or more probability intervals associated with an atomic formulae in a knowledge base to provide an interval conditional probability indicating that a first predicate condition is true based one or more alternative predicates being true.

Abstract: A purpose of the present invention is to improve precision when estimating the solution to a task expressed in a graph. A model generation device according to one embodiment of the present invention performs machine learning of an estimation module provided with a feature-incorporated network and an estimator. The feature-incorporated network is provided with a plurality of feature-incorporated layers. An encoder for each feature-incorporated layer is configured so as to incorporate the feature amount of the outflow edge and inflow edge of each vertex after reflecting the relationship with other vertices. Moreover, the encoder for each feature-incorporated layer is configured so as to derive the relative feature amount for each edge based on the feature amount for all of the edges inputted for a target vertex, without calculating the weighted sum of the features of the edges adjacent to each vertex.

Abstract: A system described herein may provide a technique for enhanced federated learning in an environment that makes use of one or more centralized models. Different nodes may be associated with different groups. Each node may provide refinement information for a given centralized model. The modifications for particular groups may be aggregated and the model may be modified based on modifications associated with each group, as opposed to modifications associated with each node. Weights for each group may be determined based on attributes of the modifications associated with each group, which may allow for the identification, on a group basis, of bias, maliciously injected data, outliers, and/or other types of modifications which may reduce the quality of the model. As such, embodiments described herein may enhance the quality, accuracy, and predictive ability of federated learning techniques that utilize distributed or federated modifications to a centralized model.

Abstract: Hierarchical structured sparse parameter pruning and processing improves runtime performance and energy efficiency of neural networks. In contrast with conventional (non-structured) pruning which allows for any distribution of the non-zero values within a matrix that achieves the desired sparsity degree (e.g., 50%) and is consequently difficult to accelerate, structured hierarchical sparsity requires each multi-element unit at the coarsest granularity of the hierarchy to be pruned to the desired sparsity degree. The global desired sparsity degree is a function of the per-level sparsity degrees. Distribution of non-zero values within each multi-element unit is constrained according to the per-level sparsity degree at the particular level of the hierarchy. Each level of the hierarchy may be associated with a hardware (e.g., logic or circuit) structure that can be enabled or disabled according to the per-level sparsity.

Abstract: Provided is a method of operating a neuron in a neuromorphic system. The method includes evaluating a membrane potential value at a corresponding time when receiving an input spike, time-modulating a synaptic weight of the membrane potential value and converting the time-modulated synaptic weight into a membrane potential value at a reference time, and generating an output spike when the membrane potential value at the reference time exceeds a certain threshold value. The membrane potential value at the reference time is represented by a floating point number including a predetermined bit of exponent and mantissa, and the floating point number includes time information. The method further includes accessing a memory and scanning a neural state variable when a timer is updated to “0” to update the neural state variable to an updated value at a reference time.

Abstract: This disclosure relates generally to a system and method for molecular property prediction. The conventional methods for molecular property prediction suffer from inherent limitation to effectively encapsulate the characteristics of the molecular graph. Moreover, the known methods are computationally intensive, thereby leading to non-performance in real-time scenarios. The disclosed method includes performing self-attention on the nodes of a molecular graph of different sized neighborhood, and further performing a shared attention mechanism across the nodes of the molecular graphs to compute attention coefficients using and Edge-condition graph attention neural network (EC-GAT). The EC-GAT effectively utilizes the edge characteristics in the molecular graph for molecular property prediction.

Abstract: An inference execution method includes: selecting an inference neural network from among a plurality of inference neural network candidates generated from one training neural network that has been trained; sequentially obtaining data; sequentially executing, on the data sequentially obtained, inference using the inference neural network; sequentially outputting results of the inference sequentially executed; and selecting a new inference neural network from among the plurality of inference neural network candidates and switching the inference neural network to be used in the execution of the inference to the new inference neural network during an inference execution period in which the data is sequentially obtained, the inference is sequentially executed, and the results of the inference are sequentially output.