LOADING...

DYNAMIC CO-ATTENTION

NETWORK

Several deep learning models have been proposed for question answering. However, due to their single-pass nature, they have no way to recover from local maxima corresponding to incorrect answers. Dynamic Coattention Network (DCN) addresses this problem for question answering. The DCN first fuses co-dependent representations of the question and the document in order to focus on relevant parts of both and then a dynamic pointer decoder iterates over potential answer spans

GO TO OVERVIEW

ABOUT

TECHNOLOGY

Majority of the world's information is recorded in the form of natural language text. Question Answering (QA) on such text is a crucial task in NLP that required both natural language understanding and world knowledge. With recent developments in deep learning, neural network models have shown promise for QA. Although these systems generally involve a smaller learning pipeline, they require a significant amount of training. GRU and LSTM units allow recurrent neural networks (RNNs) to handle the longer texts required for QA. Further improvements – such as attention mechanisms and memory networks – allow the network to focus on the most relevant facts. Such networks provide the current state-of-the-art performance for deep-learning-based QA. Dynamic Coattention Network (DCN) is an end-to-end neural network for question answering. The model consists of a coattentive encoder that captures the interactions between the question and the document, as well as a dynamic pointing decoder that alternates between estimating the start and end of the answer span.

BUSINESS

USE CASES

AI assisted Customer Service

While Artificial Intelligence technology can’t replace humans when it comes to comprehensive customer service, AI is currently being deployed in customer service to both augment human agents, improve customer experience and reduce human customer service costs.

AI assisted Legal Document Parsing

For lawyers, going through tons of legal documents like case files & legal briefs for a particular case is a cumbersome process. AI comes handy in comprehending and extracting the right information hence reducing human efforts & time and allowing them to handle multiple cases at the same time.

ARCHITECTURE

A coattention mechanism attends to the question and document simultaneously and finally fuses both attention contexts. First the affinity matrix is computed, which contains affinity scores corresponding to all pairs of document words and question words. The affinity matrix is normalized row-wise to produce the attention weights across the document for each word in the question, and column-wise to produce the attention weights across the question for each word in the document. Next, the summaries or attention contexts of the document is computed in light of each word of the question. Similarly the summaries of the question in light of each word of the document is computed. These two operations is performed in parallel. The last step is the fusion of temporal information to the coattention context via a bidirectional LSTM to produce the final coattention encoding. The encoding is unrolled by a decoder LSTM iteratively to produce a start index and end index of the answer span. A separate HMN network is used to infer the start and end index at each iteration.

Architecture

RAZORTHINK AI

DL MODEL

DESIGNER

Build this in Big Brain in just 4 steps within 30 minutes:

1Click on DL Model Designer and Drag & Drop Layers/ Operations
2Configure the parameters and so and so..
3Validate the model architecture
4Train it for your datasets.
Builder Parameters

    STATS

    1400+PARAMETERS
    12 HOURSTIME TO TRAIN ( GEFORCE GTX TITAN X )
    85,000 DOCUMENTSTRAINING DATA
    Tensorflow logo
    DESIGN TIME IN VANILLA TENSORFLOW5 - 7 DAYS
    Razorthink Logo
    DESIGN TIME IN RAZORTHINK AI12 HOURS