International Business Machines Corporation

Advanced Micro Devices Corporation

Arun C Ramachandran

Lakshmanan Velusamy

Prakash Sathyanath Raghavendra

Chandra Kumar Ramasamy

Keerthan Subraya Shagrithaya

Subramaniyam Chandrasekaran

Puvichakravarthy Ramachandran

Shunmugam Murugan

Savitri Hunasheekatti

Vasanthakumar Rajagopal

Divakar Adari

Vinodh Kumar Venkatesan

Systems, methods, and devices for pruning a convolutional neural network (CNN). A subset of layers of the CNN is chosen, and for each layer of the subset of layers, how salient each filter in the layer is to an output of the CNN is determined, a subset of the filters in the layer is determined based on the salience of each filter in the layer, and the subset of filters in the layer is pruned. In some implementations, the layers of the subset of layers of the CNN are non-contiguous. In some implementations, the subset of layers includes odd numbered layers of the CNN and excludes even numbered layers of the CNN. In some implementations, the subset of layers includes even numbered layers of the CNN and excludes odd numbered layers of the CNN.

Accelerating neural networks with one shot skip layer pruning

Systems, apparatuses, and methods for implementing one-sided per-kernel clipping and weight transformation for neural networks are disclosed. Various parameters of a neural network are quantized from higher-bit representations to lower-bit representations to reduce memory utilization and power consumption. To exploit the effective range of quantized representations, positively biased weights are clipped and negated before convolution. Then, the results are rescaled back after convolution. A one-sided clipping technique is used for transforming weights to exploit the quantization range effectively, with the side chosen to be clipped being the biased side. This technique uses a global strategy for clipping without requiring skilled expertise. This approach allows the system to retain as much information as possible without losing unnecessary accuracy when quantizing parameters from higher-bit representations to lower-bit representations.

Efficient weight clipping for neural networks

A confidential information exchange between a sender and a receiver may be conducted without the use of encryption keys. The information is coded with a Challenge-Response Table that is shared between the sender and the receiver. Rather than sending a challenge and then waiting for a response, the challenge and response are both sent by the sender of the information. The information sent comprises an index with a challenge and a response from the Challenge-Response Table. Upon receiving the coded information, the receiver uses the Challenge-Response Table to decode the information by using the index to locate the challenge and its valid response. Upon determining that the challenge and the response are correct, a first decoded answer is determined. Upon determining that either the challenge or the response, or both, are incorrect, a second decoded answer is determined.

Keyless challenge and response system

The disclosed technology generates two relative prime numbers and, then, using the relative prime numbers converts a super-increasing (SI) knapsack into a non-super-increasing (NSI) knapsack. The NSI knapsack becomes a public key and the corresponding SI knapsack, along with the two relative prime numbers, becomes a private key. A message is encrypted using a subset S of the private key that totals a number N. The message, the public key and the number N are transmitted to a recipient, who knows the value of the two relative prime numbers. The recipient uses the relative prime numbers to convert the public key into the private key and, then, generates the subset S by solving the private key with respect to the number N. Using the subset, the message is decrypted.

Generation of relative prime numbers for use in cryptography

The disclosed technology generates two relative prime numbers and, then, using the relative prime numbers converts a super-increasing (SI) knapsack into a non-super increasing (NSI) knapsack. The NSI knapsack becomes a public key and the corresponding SI knapsack, along with the two relative prime numbers, becomes a private key. A message is encrypted using a subset S of the private key that totals a number N. The message, the public key and the number N are transmitted to a recipient, who knows the value of the two relative prime numbers. The recipient uses the relative prime numbers to convert the public key into the private key and, then, generates the subset S by solving the private key with respect to the number N. Using the subset, the message is decrypted.

Disclosed herein are methods, systems, and computer program products for managing data within a file system. In embodiments, the method includes determining the location of data within a file to be modified, modifying the data within the file, and adjusting a pointer to point to the data block containing the modified data. The data within a file may be modified by deleting data from the file or inserting new data into the file. New data may be inserted into a file by writing the data into new data blocks and then adjusting the pointers to point to the location of the new data blocks. An example method may also include determining whether data blocks are incomplete blocks and identifying the incomplete blocks by negating the value of the pointers to those blocks.

Modification of data within a file

The invention includes a method for encrypting and transmitting a message M from a first location to a second location. Random numbers P and Q are initially selected, each of them being either a prime or a non-prime number. P and Q are both furnished to a sender and a recipient at the first and second locations, respectively. A client at the first location generates an encrypted message E, which is the remainder left when a modulus operator divides a first quantity, equal to M multiplied by P to the i power, where i is a selected number, by a second quantity equal to (PQ&#x2212;1). A tuple comprising E and i is transmitted from the first location to the second location, and a second data processing system at the second location to determines the message M from E, i and Q.

Use of a private key to encrypt and decrypt a message

An Eroot unit is configured to compute an Eroot of a number A with a machine. The Eroot unit locates a value E, which is even, in a first entry of a plurality of entries of a structure. The Eroot unit is configured to read a product of the pair of prime numbers from the first entry. The Eroot unit is configured to read a second of the pair of key values in the first entry. The Eroot unit is configured to encrypt the number A using the product of the pair of prime numbers to generate an encrypted value. The Eroot unit is configured to apply a decryption operation to the encrypted value using the second of the pair of key values to generate an intermediate value. The Eroot unit is configured to compute a square root of the intermediate value.

Growing community of inventors

Bengaluru, India

Arun C Ramachandran