Abstract: This disclosure relates generally to visio-linguistic understanding. Conventional methods use contextual visio-linguistic reasoner for visio-linguistic understanding which requires more compute power and large amount of pre-training data. Embodiments of the present disclosure provide a method for visio-linguistic understanding using contextual language model reasoner. The method converts the visual information of an input image into a format that the contextual language model reasoner understands and accepts for a downstream task. The method utilizes the image captions and confidence score associated with the image captions along with a knowledge graph to obtain a combined input in a format compatible with the contextual language model reasoner. Contextual embeddings corresponding to the downstream task is obtained using the combined input. The disclosed method is used to solve several downstream tasks such as scene understanding, visual question answering, visual common-sense reasoning and so on.

Abstract: Methods, systems, and computer-readable media for augmenting storage functionality using emulation of storage characteristics are disclosed. An access request for a data set is received. The access request is formatted according to a first protocol associated with a first data store, and the first data store is associated with first storage characteristics. The access request is translated into a translated access request. The translated access request is formatted according to a second protocol associated with a second data store, and the second data store is associated with second storage characteristics differing at least in part from the first storage characteristics. The translated access request is sent to the second data store. The translated access request is performed by the second data store on the data set using emulation of one or more of the first storage characteristics not included in the second storage characteristics.

Abstract: This disclosure relates generally to visio-linguistic understanding. Conventional methods use contextual visio-linguistic reasoner for visio-linguistic understanding which requires more compute power and large amount of pre-training data. Embodiments of the present disclosure provide a method for visio-linguistic understanding using contextual language model reasoner. The method converts the visual information of an input image into a format that the contextual language model reasoner understands and accepts for a downstream task. The method utilizes the image captions and confidence score associated with the image captions along with a knowledge graph to obtain a combined input in a format compatible with the contextual language model reasoner. Contextual embeddings corresponding to the downstream task is obtained using the combined input. The disclosed method is used to solve several downstream tasks such as scene understanding, visual question answering, visual common-sense reasoning and so on.

Abstract: Methods, systems, and computer-readable media for augmenting storage functionality using emulation of storage characteristics are disclosed. An access request for a data set is received. The access request is formatted according to a first protocol associated with a first data store, and the first data store is associated with first storage characteristics. The access request is translated into a translated access request. The translated access request is formatted according to a second protocol associated with a second data store, and the second data store is associated with second storage characteristics differing at least in part from the first storage characteristics. The translated access request is sent to the second data store. The translated access request is performed by the second data store on the data set using emulation of one or more of the first storage characteristics not included in the second storage characteristics.

Abstract: The online shopping is highly based on human perception on products and the human perception on products depends on semantic features of products. Conventional methods provides product recommendation based on historical data and are supervised. The present disclosure receives a set of multi-modal data. A plurality of features are extracted from the set of data at a plurality of resolution levels and the plurality of features are arranged as parallel corpus based on a category associated with each data from the set of data. Further, an abstract interaction vector is computed for each element of the set of data using the parallel corpus. Further, the set of recommendations are identified by comparing the abstract interaction vector associated with the set of data with an abstract interaction vector associated with each of a plurality of items stored in the database by utilizing a similarity metric.

Abstract: Methods, systems, and computer-readable media for record-by-record live migration using segmentation are disclosed. Migration of a data set comprises, for a record in a segment being migrated, storing a first status indicating that the record is offline in a source data store. An instance of the record is stored in the destination data store, and a second status is stored to indicate that the record is online in the destination. The record is deleted from the source after the second status is stored. During the migration, a read request for the record is received and determined to be associated with the segment being migrated. A response to the read request is generated that comprises an authoritative instance of the record. The instance of the record in the destination is determined to represent the authoritative instance based (at least in part) on the first status and the second status.

Abstract: Methods, systems, and computer-readable media for augmenting storage functionality using emulation of storage characteristics are disclosed. An access request for a data set is received. The access request is formatted according to a first protocol associated with a first data store, and the first data store is associated with first storage characteristics. The access request is translated into a translated access request. The translated access request is formatted according to a second protocol associated with a second data store, and the second data store is associated with second storage characteristics differing at least in part from the first storage characteristics. The translated access request is sent to the second data store. The translated access request is performed by the second data store on the data set using emulation of one or more of the first storage characteristics not included in the second storage characteristics.

Abstract: Methods, systems, and computer-readable media for segmentation of maintenance on distributed systems are disclosed. A data set is partitioned according to a hash function into a plurality of segments. A maintenance activity is initiated on a first segment. During the maintenance activity, a first request to perform a first action on the data set is received. Based at least in part on determining that the first request is associated with the first segment using the hash function, the first action is performed using additional processing associated with the maintenance activity. During the maintenance activity, a second request to perform a second action on the data set is received. Based at least in part on determining that the second request is associated with a second segment using the hash function, the second action is performed without the additional processing associated with the maintenance activity.

Abstract: This disclosure relates generally to quality evaluation of collaborative text input, and more particularly to system and method for quality evaluation of collaborative text inputs using Long Short Term Memory (LSTM) networks. In one embodiment, the method includes receiving an input data associated with a task to be accomplished collaboratively and sequentially by a plurality of contributors. The input data includes task-wise data sequence of contributor's post-edit submissions. A plurality of features are extracted from the input data. Based on the plurality of features, a plurality of input sequences are constructed. The input sequences include a plurality of concatenated feature vectors, where each of the concatenated feature vectors includes a post-edit feature vector and a contributor representation feature vector. The input sequences are modelled as a LSTM network, where the LSTM network is utilized to train a binary classifier for quality evaluation of the post-edit submission.

Abstract: The disclosure generally relates to transcription of spoken words, and more particularly to a system and method for transcription of spoken words using multilingual mismatched words. The process comprises collection of multi-scripted noisy transcriptions of the spoken word obtained from workers of the multilingual mismatched crowd unfamiliar with the spoken language. The collected words are mapped to a phoneme sequence in the source language using script specific graphemes to phoneme model. Further, it builds a multi-scripted transcription script specific, worker specific and a global insertion-deletion-substitution (IDS) channel. Furthermore, the disclosure also determines reputation of workers to allocate the transcription task. Determination of reputation is based on word belief.

Abstract: This disclosure relates generally to quality evaluation of collaborative text input, and more particularly to system and method for quality evaluation of collaborative text inputs using Long Short Term Memory (LSTM) networks. In one embodiment, the method includes receiving an input data associated with a task to be accomplished collaboratively and sequentially by a plurality of contributors. The input data includes task-wise data sequence of contributor's post-edit submissions. A plurality of features are extracted from the input data. Based on the plurality of features, a plurality of input sequences are constructed. The input sequences include a plurality of concatenated feature vectors, where each of the concatenated feature vectors includes a post-edit feature vector and a contributor representation feature vector. The input sequences are modelled as a LSTM network, where the LSTM network is utilized to train a binary classifier for quality evaluation of the post-edit submission.

Abstract: System and method for layer-wise training of deep neural networks (DNNs) are disclosed. In an embodiment, multiple labelled images are received at a layer of multiple layers of a DNN. Further, the labelled images are pre-processed. The pre-processed images are then transformed based on a predetermined weight matrix to obtain feature representation of the pre-processed images at the layer, the feature representation comprise feature vectors and associated labels. Furthermore, kernel similarity between the feature vectors is determined based on a predefined kernel function. Moreover, a Gaussian kernel matrix is determined based on the kernel similarity. In addition, an error function is computed based on the predetermined weight matrix and the Gaussian kernel matrix. Also, a weight matrix associated with the layer is computed based on the error function and predetermined weight matrix, thereby training the layer of the multiple layers.

Abstract: The present application provides a method and system for unsupervised word image clustering, comprises capturing one or more image wherein the one or more image comprises at least one word images. Extracting at least one feature vector using an untrained convolution neural network architecture, wherein the convolution filters are initialized by random filter based deep learning techniques using Gaussian random variable with zero mean and unit standard deviation, and wherein the convolution filters are constrained to sum to zero. The extracted feature vectors are used for clustering, wherein clustering is performed in two stages. First stage includes clustering word images which are similar using a graph connected component. Second stage clustering includes clustering a remaining word images which are not clustered during the first stage by evaluating the remaining images against the clusters formed during the first stage and assigning them to clusters based on the evaluation.

Abstract: System and method for layer-wise training of deep neural networks (DNNs) are disclosed. In an embodiment, multiple labelled images are received at a layer of multiple layers of a DNN. Further, the labelled images are pre-processed. The pre-processed images are then transformed based on a predetermined weight matrix to obtain feature representation of the pre-processed images at the layer, the feature representation comprise feature vectors and associated labels. Furthermore, kernel similarity between the feature vectors is determined based on a predefined kernel function. Moreover, a Gaussian kernel matrix is determined based on the kernel similarity. In addition, an error function is computed based on the predetermined weight matrix and the Gaussian kernel matrix. Also, a weight matrix associated with the layer is computed based on the error function and predetermined weight matrix, thereby training the layer of the multiple layers.

Abstract: The disclosure generally relates to transcription of spoken words, and more particularly to a system and method for transcription of spoken words using multilingual mismatched words. The process comprises collection of multi-scripted noisy transcriptions of the spoken word obtained from workers of the multilingual mismatched crowd. The collected words are mapped to a phoneme sequence in the source language using script specific graphemes to phoneme model. Further, it builds a multi-scripted transcription script specific, worker specific and a global insertion-deletion-substitution (IDS) channel. Furthermore, the disclosure also determines reputation of workers to allocate the transcription task. Determination of reputation is based on word belief.

Abstract: Optimizing task allocation requires taking into account cognitive load on workers and their response time to allocated tasks. The present disclosure provides for allocation of task by receiving data pertaining to current activity of workers; receiving data pertaining to at least one task to be allocated and determining activity-task pairs based on an activity feature vector corresponding to at least one human body part used during the current activity and a task feature vector corresponding to at least one human body part required for the at least one task to be performed by the workers. Cognitive load on the workers is then estimated for the determined activity-task pairs. An optimum activity-task pair based on the estimated cognitive load is determined and at least one task is allocated to the workers based on the determined optimum activity-task pair.

Abstract: The present subject matter discloses a system and a method for allocating task on crowdsourcing platform. A task may be received from a first user on the platform. Further, a protocol may be configured by the first user indicating one or more task actions to be performed for completing the task. Further, a hierarchy comprising a plurality of micro-tasks associated with the task may be created. Based on the protocol configured, the system may assign a task action to each of the plurality of micro-tasks. Further, the system may allocate each of the plurality of micro-tasks to a second user based on the task action assigned and a set of parameters. The set of parameters may comprise second user's metadata, completion time associated with each micro-task, size of each micro-task, and form factor of an interface accessed by the second user.