ISE Developer Blog

Machine Learning - ISE Developer Blog

Making sense of Handwritten Sections in Scanned Documents using the Azure ML Package for Computer Vision and Azure Cognitive Services

May 7, 2018 May 7, 2018 05/7/18
Tingting Zhao
Extracting general concepts, rather than specific phrases, from documents and contracts is challenging. It's even more complicated when applied to scanned documents containing handwritten annotations. We describe using object detection and OCR with Azure ML Package for Computer Vision and Cognitive Services API.

Sequence Intent Classification Using Hierarchical Attention Networks

March 6, 2018 Mar 6, 2018 03/6/18
Olga Liakhovich
We analyze how Hierarchical Attention Neural Networks could be helpful with malware detection and classification scenarios, demonstrating the usefulness of this approach for generic sequence intent analysis. The novelty of our approach is in applying techniques that are used to discover structure in a narrative text to data that describes the behavior of executables.

Permissively-Licensed Named Entity Recognition on the JVM

November 20, 2017 Nov 20, 2017 11/20/17
Clemens Wolff
The ability to correctly identify entities, such as places, people, and organizations, adds a powerful level of natural language understanding to applications. This post introduces a MIT-licensed one-click deployment to Azure for web services that lets developers get started with a wide range of natural language tasks in 5 minutes or less, by consuming simple HTTP services for language identification, tokenization, part-of-speech-tagging and named entity recognition.

Building a Custom Spark Connector for Near Real-Time Speech-to-Text Transcription

November 1, 2017 Nov 1, 2017 11/1/17
Clemens Wolff
This post describes in detail the Azure Cognitive Services speech-to-text WebSocket protocol and shows how to implement the protocol in Java. This enables us to transcribe audio to text in near real-time. We then show how to feed the transcribed radio into a pipeline based on Spark Streaming for further analysis, augmentation, and aggregation. The Java client is reusable across a wide range of text-to-speech scenarios that require time-efficient speech-to-text transcription in more than 10 languages including English, French, Spanish, German and Chinese.