Abstract: A system for facilitating offline processing of electronic transactions at an ATM when the ATM does not support a network connection may be provided. The system may include a mobile device configured to store an updatable account balance of a personal account associated with a user of the mobile device and transaction data associated with a plurality of transactions executed via the mobile device. The system may include the ATM being configured to operate in an offline processing mode when the network connection is unsupported. The offline processing mode may include the ATM receiving a transaction request and account data from the mobile device. The ATM may be configured to authenticate or deny the executing of the transaction request based on a predetermined threshold of transaction locations positioned at a distance greater than or less than a pre-determined distance from the ATM.

Abstract: A method using one or more mobile devices for transporting information from an automated teller machine (“ATM”) to a central server when the ATM does not support a network connection and, upon return of the one or more mobile devices to the ATM, updating the ATM may be provided. The method may include transporting electronic transaction data processed locally at the ATM, via a mobile device, to a location where a network connection between the mobile device and the central server may be established. When in a location including network connection, the method may include transmitting the electronic transaction data to the central server. When the mobile device is detected to have returned to be within the pre-determined range of the ATM, the method may include transmitting a data packet to the ATM, the data packet received from the central server, thereby updating the ATM of the transmittal.

Abstract: Methods, systems, and apparatus, including computer-readable media, are described for performing neural network computations using a system configured to implement a neural network on a hardware circuit. The system includes a host that receives a batch of inputs to a neural network layer. Each of the inputs is stored in a memory location identified by an address. The system identifies one or more duplicate addresses in a listing of addresses for one or more inputs. For each duplicate address: the system generates a unique identifier that identifies the duplicate address in the listing of addresses. The system (i) obtains first inputs from memory locations identified by addresses corresponding to the unique identifiers and (ii) generates an output of the layer from the obtained first inputs.

Abstract: A method is provided to reduce the number of duplicates of each document that is stored within entity databases. The method may include creating discrete links and/or pointers to the location of the document already stored within an entity. The method may also include separating the document into different classification levels. The method may include storing the different parts of the documents in different locations within the entity.

Abstract: Methods, systems, and apparatus, including computer-readable media, are described for performing neural network computations using a system configured to implement a neural network on a hardware circuit. The system includes a host that receives a batch of inputs to a neural network layer. Each of the inputs is stored in a memory location identified by an address. The system identifies one or more duplicate addresses in a listing of addresses for one or more inputs. For each duplicate address: the system generates a unique identifier that identifies the duplicate address in the listing of addresses. The system (i) obtains first inputs from memory locations identified by addresses corresponding to the unique identifiers and (ii) generates an output of the layer from the obtained first inputs.

Abstract: Methods, systems, and apparatus, including computer-readable media, are described for performing neural network computations using a system configured to implement a neural network on a hardware circuit. The system includes a host that receives a batch of inputs to a neural network layer. Each of the inputs is stored in a memory location identified by an address. The system identifies one or more duplicate addresses in a listing of addresses for one or more inputs. For each duplicate address: the system generates a unique identifier that identifies the duplicate address in the listing of addresses. The system (i) obtains first inputs from memory locations identified by addresses corresponding to the unique identifiers and (ii) generates an output of the layer from the obtained first inputs.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using dynamic minibatch sizes during neural network training. One of the methods includes receiving, by each of a plurality of host computer, a respective batch of training examples, each training example having zero or more features, computing, by each host computer, a minimum number of minibatches into which the host computer can divide the respective batch of training examples so that the host computer can process each minibatch using an embedding layer of the neural network without exceeding available computing resources, determining a largest minimum number of minibatches (N) into which any host computer can divide its respective batch of training examples, generating, by each host computer, N minibatches from the respective batch of training examples received by the host computer, and processing, by each host computer, the N minibatches using the embedding layer.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using dynamic minibatch sizes during neural network training. One of the methods includes receiving, by each of a plurality of host computer, a respective batch of training examples, each training example having zero or more features, computing, by each host computer, a minimum number of minibatches into which the host computer can divide the respective batch of training examples so that the host computer can process each minibatch using an embedding layer of the neural network without exceeding available computing resources, determining a largest minimum number of minibatches (N) into which any host computer can divide its respective batch of training examples, generating, by each host computer, N minibatches from the respective batch of training examples received by the host computer, and processing, by each host computer, the N minibatches using the embedding layer.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using dynamic minibatch sizes during neural network training. One of the methods includes receiving, by each of a plurality of host computer, a respective batch of training examples, each training example having zero or more features, computing, by each host computer, a minimum number of minibatches into which the host computer can divide the respective batch of training examples so that the host computer can process each minibatch using an embedding layer of the neural network without exceeding available computing resources, determining a largest minimum number of minibatches (N) into which any host computer can divide its respective batch of training examples, generating, by each host computer, N minibatches from the respective batch of training examples received by the host computer, and processing, by each host computer, the N minibatches using the embedding layer.