際際滷

際際滷Share a Scribd company logo

Building Heuristic ARchitecture for Artificial InTelligence (BHARAT) NLP with NITI Aayog

Ekta Grover
Ekta Grover

I had the opportunity to represent Inmobi at BHARAT NLP workshop, which is a NITI Aayog initiative towards the National Strategy for Artificial Intelligence in India. NITI is a policy think tank of the Government of India which aims at sustainable development goals. The goal of the workshop was to architect a single access point repository of NLP tools for Indic languages. This would lower the entry barrier for delivery of dynamic content in Indic languages and in-turn make the products and services accessible in all Indic languages and their user-base. To facilitate this, we had interactive sessions with leading academics, industry experts, and startups working across NLP use-cases - which finally culminated into each of us breaking out in group discussions, coming up with the design - and presenting our architecture & insights. This work is the sum total of our learnings as a group.

1 of 3
Download to read offline
Team : (alphabetical order) AG Ramakrishnan (Indian Institute of Science) Anurag Bajpai (Shadowfax) Ekta Grover (Inmobi) Kalika Bali (Microsoft Research) Kartik Ukhalkar (Udaan) Raghavendra Bhat (Intel) Dr. Rajeev RR ((ICFOSS) Sumant Vidwans (Facebook) Sunny Manchanda (DRDO) Prof. Uma Maheshwari (University of Hyderabad) Aalekh Sharan /Avik Sarkar/Rishi Razdan - facilitators , NITI Aayog Building Heuristic ARchitecture for Artificial InTelligence (BHARAT) NLP 30th November, 2018 Microsoft Research Labs, Bangalore
Contextual, application aware accuracy tiers and operability support for customer tiers Allow graceful degradation across customer tiers Performance (Availability, Latency, Throughput Guarantees & other operability guarantees) Goals Open Community Focus Inclusive of Consumer + Producer use-cases Not prescriptive - a la-carte Power variety of content use-cases Customer Tiers Vendor Neutral SLA Goals for an Enterprise grade Platform
Licensing Data Access Platform Toolkits Voice Biometrics DataCollection Apps Static/Dynamic Curated/Crowdsourced Centralized/Distributed Domain/Semantics Authentication Audit Data Pipelines OCR TTS NLU ASR Authorization HWR Utilities Services Indexing Search Chat Bots Context Switching Experimentation Representation Standards Evaluation Model Certification / SLA Tunable Accuracy / Relevance Feedback Custom / Legacy Data Sets Decision Function Custom Plugins Language Connotations Lossless Translation Apriori Learning Baseline Models Recipes / Cookbooks Transfer Learning Localization Task Based Understanding Synonyms Sparse Data Models Standard Data Sets Locale Data Sets Translation Geocoding Multimodal Content Consumption Analytics Multilingual Interoperability Intent Summarization POSParsers chunking Payment plugins Gateways integration Key vault Custom model plugins Language Models OperabilitySupport

More Related Content

Building Heuristic ARchitecture for Artificial InTelligence (BHARAT) NLP with NITI Aayog

  • 1. Team : (alphabetical order) AG Ramakrishnan (Indian Institute of Science) Anurag Bajpai (Shadowfax) Ekta Grover (Inmobi) Kalika Bali (Microsoft Research) Kartik Ukhalkar (Udaan) Raghavendra Bhat (Intel) Dr. Rajeev RR ((ICFOSS) Sumant Vidwans (Facebook) Sunny Manchanda (DRDO) Prof. Uma Maheshwari (University of Hyderabad) Aalekh Sharan /Avik Sarkar/Rishi Razdan - facilitators , NITI Aayog Building Heuristic ARchitecture for Artificial InTelligence (BHARAT) NLP 30th November, 2018 Microsoft Research Labs, Bangalore
  • 2. Contextual, application aware accuracy tiers and operability support for customer tiers Allow graceful degradation across customer tiers Performance (Availability, Latency, Throughput Guarantees & other operability guarantees) Goals Open Community Focus Inclusive of Consumer + Producer use-cases Not prescriptive - a la-carte Power variety of content use-cases Customer Tiers Vendor Neutral SLA Goals for an Enterprise grade Platform
  • 3. Licensing Data Access Platform Toolkits Voice Biometrics DataCollection Apps Static/Dynamic Curated/Crowdsourced Centralized/Distributed Domain/Semantics Authentication Audit Data Pipelines OCR TTS NLU ASR Authorization HWR Utilities Services Indexing Search Chat Bots Context Switching Experimentation Representation Standards Evaluation Model Certification / SLA Tunable Accuracy / Relevance Feedback Custom / Legacy Data Sets Decision Function Custom Plugins Language Connotations Lossless Translation Apriori Learning Baseline Models Recipes / Cookbooks Transfer Learning Localization Task Based Understanding Synonyms Sparse Data Models Standard Data Sets Locale Data Sets Translation Geocoding Multimodal Content Consumption Analytics Multilingual Interoperability Intent Summarization POSParsers chunking Payment plugins Gateways integration Key vault Custom model plugins Language Models OperabilitySupport