SEMINAR REPORT_PUJITYG_1JS17CS073

VISVESVARAYA TECHNOLOGICAL UNIVERSITY Belagavi-590 018, KARNATAKA.

A SEMINAR REPORT ON

“NATIONAL STRATEGY FOR AI” Submitted in partial fulfilment of the requirements for the Seminar(17CSS86) course of 8th semester

BACHELOR OF ENGINEERING IN COMPUTER SCIENCE AND ENGINEERING Submitted by Pujit Y G (1JS17CS073)

Under the guidance of Ms. Pooja H, B.E., M.Tech Assistant Professor Department of Computer Science and Engineering JSSATE, Bengaluru

JSS ACADEMY OF TECHNICAL EDUCATION-BENGALURU Department of Computer Science and Engineering 2020 – 2021

JSS MAHAVIDYAPEETHA, MYSURU JSS ACADEMY OF TECHNICAL EDUCATION JSS Campus, Uttarahalli-Kengeri Main Road, Bengaluru - 560060

Department of Computer Science and Engineering

CERTIFICATE This is to Certify that seminar work entitled “NATIONAL STRATEGY FOR AI” has been successfully completed by Pujit YG, USN: 1JS17CS073, a bonafide student of JSS ACADEMY OF TECHNICAL EDUCATION, BENGALURU in partial fulfilment of the requirements of the 8th semester for award of degree in Bachelor of Engineering in Computer Science and Engineering of Visvesvaraya Technological University, Belagavi during academic year 2020-2021. The seminar report has been approved as it satisfies the academic requirements in respect of seminar work for the said degree.

Ms. Pooja H Assistant Professor Department of CSE JSSATE, Bengaluru

Dr. Naveen N C Professor and Head Department of CSE JSSATE, Bengaluru

Name of the Examiners

Signature with date

1)………………………….

…………………………

2)………………………….

…………………………

ACKNOWLEDGEMENT I express my humble pranamas to His Holiness Jagadguru Sri Sri Sri Shivarathri Deshikendra Mahaswamiji who has showered their blessings on us for framing our career successfully. The fulfilment and rapture that go with the fruitful finishing of any assignment would be inadequate without the specifying the people who made it conceivable, whose steady direction and support delegated the endeavours with success. I would like to profoundly thank Management of JSS ACADEMY OF TECHNICAL EDUCATION-BENGALURU for providing such a healthy environment to carry out this Seminar work. I take immense pleasure in thanking Dr. Mrityunjaya V Latte, Principal, JSSATE, Bengaluru, for being kind enough to provide us with an opportunity to work the Project in this institution. I’m also thankful to Dr Naveen N C, Professor and Head of Department of Computer Science and Engineering, for his co-operation and encouragement at all moments of approach. I’m also thankful to our Seminar guide Mrs. Pooja H, Assistant Professor, for her constant support and encouragement. I’m also thankful to Mr. Renuka Rajendra. B. and Dr. Naidila Sadashiv, Seminar coordinators, for their cooperation and support. I wish to thank every teaching and non-teaching faculty of out department for always being there to support and guide us.

Pujit Y G 1JS17CS073

ABSTRACT Artificial Intelligence (AI) is poised to disrupt our world. With intelligent machines enabling high-level cognitive processes like thinking, perceiving, learning, problem solving and decision making, coupled with advances in data collection and aggregation, analytics and computer processing power, AI presents opportunities to complement and supplement human intelligence and enrich the way people live and work. India, being the fastest growing economy with the second largest population in the world, has a significant stake in the AI revolution. Recognising AI’s potential to transform economies and the need for India to strategize its approach with a view to guiding the research and development in new and emerging technologies

TABLE OF CONTENTS Chapter title

Page No

Abstract

i

Acknowledgement

ii

Table of contents

iii

List of figures

iv

Chapter 1: Introduction

1

1.1 Background

1

1.2 Artificial Intelligence

2

1.3 Global developments in AI

3

Chapter 2: Literature survey

5

Chapter 3: AIRAWAT

6

3.1 AIRAWAT Overview

6

3.2 AIRAWAT Introduction

7

3.3 Potential Structures for AIRAWAT

9

3.4 Architecture for AIRAWAT

10

Chapter 4: Reference model for AIRAWAT

12

4.1 Summit and ABCI

12

4.2 ABCI Innovation Platform

13

4.3 ABCI Architecture

13

Chapter 5: Research and ideas

16

5.1 Health care

16

5.2 Agriculture

17

5.2.1 AI Sowing App

18

5.3 Smart cities and Infrastructure

19

5.4 Education

20

5.4.1 Smart Content

21

5.4.2 WriteToLearn by Pearson

21

5.4.3 Predicting dropouts in Andra Pradesh

21

Chapter 6: Conclusion

22

References

23

LIST OF FIGURES Figure No.

Figure Title

Page No.

1.1

Artificial Intelligence

2

3.1

HPC vs AI Computer Infrastructure

7

3.2

AIRAWAT Architecture

10

4.1

ABCI Platform

13

4.2

ABCI Architecture

12

5.1

AI and Robotics in health

16

5.2

Accenture Research in Agriculture

18

NATIONAL STRATEGY FOR AI

Chapter 1

INTRODUCTION 1.1 Background Artificial Intelligence (AI) is poised to disrupt our world. With intelligent machines enabling high-level cognitive processes like thinking, perceiving, learning, problem solving and decision making, coupled with advances in data collection and aggregation, analytics and computer processing power, AI presents opportunities to complement and supplement human intelligence and enrich the way people live and work. Since the start of this year, NITI Aayog has partnered with several leading AI technology players to implement AI projects in critical areas such as agriculture and health. Learnings from these projects, under various stages of implementation, as well as our engagement with some of the leading institutions and experts have given a better perspective to our task of crafting the national strategy for AI, which is the focus of this discussion paper. This strategy document is premised on the proposition that India, given its strengths and characteristics, has the potential to position itself among leaders on the global AI map – with a unique brand of #AIforAll. The approach in this paper focuses on how India can leverage the transformative technologies to ensure social and inclusive growth in line with the development philosophy of the government. In addition, India should strive to replicate these solutions in other similarly placed developing countries. #AIforAll will aim at enhancing and empowering human capabilities to address the challenges of access, affordability, shortage and inconsistency of skilled expertise; effective implementation of AI initiatives to evolve scalable solutions for emerging economies; and endeavors to tackle some of the global challenges from AI’s perspective, be it application, research, development, technology, or responsible AI. #AIforAll will focus on harnessing collaborations and partnerships, and aspires to ensure prosperity for all. Thus, #AIforAll means technology leadership in AI for achieving the greater good. From an applications perspective, the approach is to identify sectors that may have the potential of greatest externalities while adopting AI solutions, and hence require the government to play a leading role in developing the implementation roadmap for AI. For example, the agriculture sector in India, which forms the bedrock of India’s economy, needs multi-layered technology infusion and coordination amongst several stakeholders. Efforts from private sector may neither be financially optimal nor efficient on a standalone basis, and hence sustained government intervention to tackle the existing challenges and constraints is needed. Hence, India’s approach to implementation of AI has to be guided by Dept of CSE, JSSATE

2020-2021

Page 1

NATIONAL STRATEGY FOR AI optimisation of social goods, rather than maximisation of top line growth. Solving for India, given the complexity and multi-dimensional aspects of most of our economic and societal challenges, can easily be extended to the rest of the emerging and developing economies. The purpose of this paper is to lay the ground work for evolving the National Strategy for Artificial Intelligence.

1.2 Artificial Intelligence AI might just be the single largest technology revolution of our live times, with the potential to disrupt almost all aspects of human existence. Andrew Ng, Co-founder of Coursera and formerly head of Baidu AI Group / Google Brain, compares the transformational impact of AI to that of electricity 100 years back. With many industries aggressively investing in cognitive and AI solutions, global investments are forecast to achieve a compound annual growth rate (CAGR) of 50.1% to reach USD57.6 billion in 2021. AI is not a new phenomenon, with much of its theoretical and technological underpinning developed over the past 70 years by computer scientists such as Alan Turing, Marvin Minsky and John McCarthy. AI has already existed to some degree in many industries and governments. Now, thanks to virtually unlimited computing power and the decreasing costs of data storage, we are on the cusp of the exponential age of AI as organisations learn to unlock the value trapped in vast volumes of data.

Figure 1.1 Artificial Intelligence AI is a constellation of technologies that enable machines to act with higher levels of intelligence and emulate the human capabilities of sense, comprehend and act. Thus, Dept of CSE, JSSATE

2020-2021

Page 2

NATIONAL STRATEGY FOR AI computer vision and audio processing can actively perceive the world around them by acquiring and processing images, sound and speech. The natural language processing and inference engines can enable AI systems to analyse and understand the information collected. An AI system can also take action through technologies such as expert systems and inference engines or undertake actions in the physical world. These human capabilities are augmented by the ability to learn from experience and keep adapting over time. AI systems are finding ever-wider application to supplement these capabilities across enterprises as they grow in sophistication. Irrespective of the type of AI being used, however, every application begins with large amounts of training data. In the past, this kind of performance was driven by rules-based data analytics programs, statistical regressions, and early “expert systems.” But the explosion of powerful deep neural networks now gives AI something a mere program doesn’t have: the ability to do the unexpected.

1.3 Global Developments in AI Countries around the world are becoming increasingly aware of the potential economic and social benefits of developing and applying AI. For example, China and U.K. estimate that 26% and 10% of their GDPs respectively in 2030 will be sourced from AI-related activities and businesses. There has been tremendous activity concerning AI policy positions and the development of an AI ecosystem in different countries over the last 18 to 24 months – the US published its AI report in December 2016; France published the AI strategy in January 2017 followed by a detailed policy document in March 2018; Japan released a document in March 2017; China published the AI strategy in July 2017; and U.K. released its industrial strategy in November 2017. Infrastructural supply side interventions have been planned by various countries for creating a larger ecosystem of AI development. Creation of “data trusts”, rolling out of digital connectivity infrastructure such as 5G / full fiber networks, common supercomputing facilities, fiscal incentives and creation of open source software libraries are some of the focus areas of various governments as committed in their strategy papers. In the area of core research in AI and related technologies, universities and research institutions from the US, China and Japan have led the publication volume on AI research topics between 2010 and 2016. Universities in USA, primarily Carnegie Mellon University, Massachusetts Institute of Technology and Stanford, took an early lead in AI research by offering new courses, establishing research facilities and instituting industry partnerships. Off late, Chinese universities, especially Peking and Tsinghua Universities have caught on Dept of CSE, JSSATE

2020-2021

Page 3

NATIONAL STRATEGY FOR AI to the race by utilising large scale public funding and extensive research partnerships with private companies. For building the future workforce for AI, countries are also significantly increasing the allocation of resources for Science, Technology, Engineering and Maths (STEM) talent development through investment in universities, mandating new courses (e.g., AI and law), and offering schemes to retrain people. For instance, U.K. has planned to build over 1,000 government supported PhD researchers by 2025 and set up a Turing fellowship to support an initial cohort of AI fellows while China has launched a five-year university program to train at least 500 teachers and 5,000 students working on AI technologies. Governance structures for enabling all the above mandates vary across countries. Many countries have instituted dedicated public offices such as Ministry of AI (UAE), and Office of AI and AI Council (U.K.) while China and Japan have allowed existing ministries to take up AI implementation in their sectoral areas. Not just national governments, but even local city governments have become increasingly aware about the importance and potential of AI and have committed public investments. National governments have significantly increased public funding for AI through commitments such as increasing the R&D spend, setting up industrial and investment funds in AI startups, investing in network and infrastructure and AI-related public procurements. China, USA, France and Japan have committed significant public spending for AI technology development and adoption. These countries are also leveraging different combinations of public-private-academia to develop and promote AI. Development of technology parks, and connecting large corporations with startups and Discussion Paper National Strategy for Artificial Intelligence 17 forming “national teams” with large private players to undertake fundamental and applied research are some of the public-private partnership approaches various national governments have espoused. AI technology development and applications are evolving rapidly with major implications for economies and societies. A study by EY and NASCCOM found that by 2022, around 46% of the workforce will be engaged in entirely new jobs that do not exist today, or will be deployed in jobs that have radically changed skillset . If some countries decide to wait for a few years to establish an AI strategy and put in place the foundations for developing the AI ecosystem, it seems unlikely that they would be able to attain and match up to the current momentum in the rapidly changing socio-economic environment. Therefore, the need of the hour is to develop a policy framework that will help set up a vibrant AI ecosystem in India. Dept of CSE, JSSATE

2020-2021

Page 4

NATIONAL STRATEGY FOR AI

Chapter 2

LITERATURE SURVEY 1. Rajat Kathuria, Mansi Kedia & Sashank Kapilavai in their paper titled “Implication of AI on Indian Economy” evaluates the growth potential and dynamics of a revolutionary technology, Artificial Intelligence, in the Indian ecosystem. In the paper they have also recommend various policy changes that could favour growth of AI in India. They described about the need to build India’s very own AI infrastructure, which assists in engagement between Government, Industry and Academia. They also provided various solution to address various governance challenges in AI. 2. Niti Aayog in their paper title “National Strategy for AI” gave an in depth means to establish and manage Artificial intelligence in various sectors. It also brought various Artificial intelligence projects, which are being developed or already developed, by companies into light. The paper tried to give overview of the cloud infrastructure for Artificial intelligence. 3. Niti Aayog in the paper titled “AIRAWAT-Establishing AI Specific Cloud Computing”, explained to need to establish India’s very own AI specific cloud infrastructure, and it also explained the architecture of the AIRAWAT. With reference to various summits, the paper described other similar architecture for Artificial Intelligence. One of the reference model explained in this paper is ABCI, a Japanese model, it helped the readers understand the cons of using the ABCI’s specification for Indian AIRAWAT, with little modification.

Dept of CSE, JSSATE

2020-2021

Page 5

NATIONAL STRATEGY FOR AI

Chapter 3

AIRAWAT 3.1 AIRAWAT OVERVIEW NITI Aayog released the National Strategy for Artificial Intelligence (NSAI) discussion paper in June 2018, in pursuance of the mandate entrusted to it by the Hon’ble Finance Minister in the Budget Speech of 2018 – 2019. NSAI highlighted the potential of Artificial Intelligence (AI) in boosting India’s annual growth rate by 1.3 percentage points by 2035 and identified priority sectors for the deployment of AI with Government’s support (Healthcare, Agriculture, Education, Smart Cities and Mobility). NSAI also emphasized on four broad recommendations in supporting and nurturing an AI ecosystem in India: (a) promotion of research (b) skilling and reskilling of the workforce (c) facilitating adoption of AI solutions; and (d) the development of guidelines for ‘responsible AI’. AIRAWAT is well in line with India’s recent approach to innovation in fields of emerging and digital technology fields. This has been an approach of facilitation of innovation, rather than implementation, where we have seen large government funding for the creation of digital infrastructure aimed at enabling research and innovation, like the creation of the Unified Payments Interface (UPI), an underlying infrastructure for payments. UPI has grown tremendously over just 4 years as multiple products and innovators have leveraged it’s capabilities and is widely credited for India’s digital payments revolution. As a computing facility designed specifically to execute tasks relevant to Machine Learning (ML) / Deep Learning (DL) applications, it is our hope that AIRAWAT will have a similar effect of bolstering AI research and application in India. AI computing infrastructure is distinct from High Performance Computing (HPC) infrastructure and the difference needs to be well understood for purposes of future infrastructure planning. HPCs, with its origins in particle physics simulations, have dominated the hardware development for several decades. “HPCs are designed by aggregating clusters of computers designed specifically for delivering higher performance (as compared to a typical desktop computer or workstation) in order to solve large problems in science, engineering, or business3”. Dept of CSE, JSSATE

2020-2021

Page 6

NATIONAL STRATEGY FOR AI The following chart captures representative difference between an HPC and a GPU-enabled AI compute infrastructure:

Figure 3.1 HPC vs AI Computer Infrastructure From a storage perspective, AI infrastructure involves very large datasets and storage transactions that are read-dominated at the beginning of each epoch (an epoch is defined as one complete pass-through of the datasets, inclusive of multiple iterations of model parameter updates). This differs from typical HPC applications which are write-intensive. ML / DL training is usually static, involving large groups of random reads, accessed repeatedly, since the same data is used for training over and over. The iterative nature of optimization of ML / DL algorithms necessitates the availability of a large amount of specialized computing resources for their continuous testing. The lack of availability of these resources is often cited as a major hurdle to the creation of a vibrant ecosystem for research in AI in India4 . It is envisaged that if made available, the specialized compute resource would not only significantly improve the outlook of research in the field in India, but also increase India’s competitiveness in international conferences and journal publications. The building of an indigenous compute facility, rather than increasing reliance on third party solutions (AWS, Azure, etc.) would also allay concerns of data privacy, while simultaneously increasing capacity to create and deploy similar facilities in India in the future.

3.2 AIRAWAT INTRODUCTION Existing and recent efforts of the Government, viz. NSAI and the National Mission for Interdisciplinary Cyber Physical Systems (NM-ICPS), have emphasized the need for Dept of CSE, JSSATE

2020-2021

Page 7

NATIONAL STRATEGY FOR AI enhancing both the core and applied research capabilities in AI, through initiatives like setting up of COREs (Centers of Research Excellence), ICTAIs (International Centers Transformational AI) and Innovation Hubs. In addition several other initiatives are being taken by governments and private sector to increase the adoption of AI, both in governance and private enterprises. These initiatives would spur the demand and necessity for state-ofthe-art and specialised AI computing infrastructure. In order to meet this demand and tackle the challenges associated with lack of access to computing resources highlighted, it is proposed that an AI-specific compute infrastructure be established. Such an infrastructure will power the computing needs of COREs, ICTAIs and Innovation Hubs, as well as facilitate the work of broader spectrum of stakeholders in the AI research and application ecosystem (startups, researchers, students, government organizations, etc.). The proposal to establish India’s own AI-first compute infrastructure is aimed to facilitate and speed up research and solution development for solving India’s societal challenges using high performance and high throughput AI-specific supercomputing technologies. The key design considerations for this infrastructure are: 1. Institutional framework for implementation: an interdisciplinary task force 2. Structure of the facility: whether it should be centralized (in a single location), decentralized (access from across multiple locations) or utilize existing infrastructure (through existing Cloud Service Providers or existing HPC infrastructure); 3. Modes of access: whether it should be made available similar to access mechanisms for a traditional HPC or through as a fully managed cloud service; 4. Architecture of facility: what would constitute the roader technical design considerations; The proposed infrastructure is acronymed AIRAWAT, i.e. the “AI Research, Analytics and knowledge Assimilation platform”) and the design suggested is in line with the recommendations of the NSAI. Given the inter-disciplinary nature of AI that would involve multiple entities, NITI Aayog recommends setting up of an inter-ministerial body (Task Force), with cross-sectoral representation, to spearhead the implementation of AIRAWAT. The Task Force may include representation of both developer community and user domain experts of this infrastructure facility, in advisory capacity, to ensure that the design of the facility is robust and is truly reflective of the demands of the stakeholders and keeps innovating with the evolving nature of technology. Dept of CSE, JSSATE

2020-2021

Page 8

NATIONAL STRATEGY FOR AI The proposed development of AIRAWAT would be in line with the approach the Government has taken of developing common public infrastructure and enabling the various stakeholders to leverage the public good to innovate and achieve the stated goals. This approach has led to India leapfrogging the world in the field of digital payments by building world’s most advanced payment system, UPI. UPI, which was developed as a public good, in partnership with 12 banks, has now more than 143 banks live on it and has registered more than USD125bn in transactions since its inception in August 2016. UPI now constitutes more than 50% of all online payments, and has raced ahead of cards and other modes to become the most preferred payment method. Developing AIRAWAT is expected to similarly invigorate the AI ecosystem in India, addressing the computing infrastructure needs of startups, academicians, researchers etc. As such, the AIRAWAT should be seen as an essential public good and funded by the Government. The necessary funding for AIRAWAT may be provided by supplementing funds under the NSM.

3.3 POTENTIAL STRUCTURES FOR AIRAWAT 1. Utilizing existing HPC infrastructure i.

Current Installed Supercomputers are designed for HPC.

ii.

Rigid to upgrade to AI workload: HPC overloaded.

iii.

New initiatives e.g. NSM are also HPC focused.

2. Creating a new decentralized facility i.

Optimal only for small R&D.

ii.

Collaboration / aggregation / workload distribution / administration challenges.

iii.

Repetitive cost.

3. Utilizing existing ‘public cloud’ infrastructure i.

Data sharing concerns.

ii.

Lack of clarity and policy on data security / privacy.

iii.

Non-predictable and high bandwidth costs.

iv.

Suitable for pay as you go and small instance requirements.

4. Creating a new centralized facility i.

No data sharing concerns.

ii.

Reuse existing high bandwidth infra (e.g. National Knowledge Network).

iii.

Efficient utilization in multi-user and multi-tenant environment.

Dept of CSE, JSSATE

2020-2021

Page 9

NATIONAL STRATEGY FOR AI iv.

Can support both small experiments as well as grand challenges / big data.

3.4 ARCHITECTURE OF AIRAWAT From a technical specification perspective, the most important aspect is building an AI infrastructure that is scalable and flexible, and can cater to rapidly changing AI development landscape. We are currently in the phase of narrow AI, defined by performance in a single domain with human or superhuman accuracy and speed for certain tasks, which have been broadly adopted in applications from facial recognition to natural language translation. We are just at the beginning of Broad AI, which encompasses multitask, multi-domain, multi-model, distributed and explainable AI. Transfer learning and reasoning are central to expanding AI to small datasets. Reducing the time and power requirements of AI computing is fundamental to the development and adoption of Broad AI solutions, and thus will dictate the technical specifications of computing infrastructure being designed. While the technical specifications for AIRAWAT will be evolved and designed through an open request for proposal process, it is recommended that the technical capabilities may be designed on the lines of the Summit and ABCI facilities.

Figure 3.2 AIRAWAT Architecture

The broad specifications that may be considered for AIRAWAT architecture may include: (a) Multi-tenant multi-user computing support. (b) Resource partitioning and provisioning, dynamic computing environment.

Dept of CSE, JSSATE

2020-2021

Page 10

NATIONAL STRATEGY FOR AI (c) ML / DL software stack – training and inferencing development kit, frameworks, libraries, cloud management software. (d) Support for varieties of AI workloads and ML / DL frameworks for user choices. (e) Energy-saving, high teraflops per watt per server rack space. (f) Low latency high bandwidth network. (g) Multi-layer storage system to ingest and process multi-petabytes of big data. (h) Compatibility with NKN (with upgrade to NKN, if needed). The proposed architecture, with composite compute and storage infrastructure allows maintaining large data sets (thus eliminating the need for separate data centres and addressing data integrity concerns), and proximity of compute facility for efficient processing of data-intensive tasks viz. training of algorithms on large (both number and size) datasets. The expected infrastructure, with capabilities of more than 100 peta flops (in the simplest sense, an AI flop is a measure of how fast a computer can perform deep neural network operations), would be more than the combined computing facility of top 20 supercomputers in India, and will put India on the global AI map, at par with the likes of Europe and Japan. Energy efficiency will be a key aspect of the facility, with the aim of putting AIRWAT in the list of top global green supercomputers. The facility would also enable storing of India’s massive data sets from areas like healthcare, agriculture locally in a high throughput and efficient storage. This new centralised AI infrastructure would alleviate any data sharing concerns (eliminating need to share data at multiple decentralised locations), is aimed at reusing existing high bandwidth infrastructure (e.g. NKN), is a better approach to utilization of computing resources in multi-user and multi-tenant environment, and has the scaling flexibility to include both small experiments as well as solving grand challenges / big data. The use cases for AIRAWAT may vary from Big Data Analytics to specialised AI solutions across multiple domains viz. Healthcare (precision diagnostics, non-invasive diagnostics etc.), Agriculture (precision agriculture, crop infestations, advanced agronomic advisory etc.), weather forecasting, security and surveillance, financial inclusion and other services (fraud detection), infrastructural tools viz. NLP etc.

Dept of CSE, JSSATE

2020-2021

Page 11

NATIONAL STRATEGY FOR AI

Chapter 4

REFERENCE MODEL FOR AIRAWAT 4.1 SUMMIT AND ABCI The Summit supercomputer at Oak Ridge National Laboratory, commissioned by the US Department of Energy, embodies multiple features of system-level purpose-built architecture for AI computation. The supercomputing facility, developed by IBM, was ranked the #1 most powerful supercomputer in the world in June 2018. The Summit architecture is designed not only for raw performance, but specifically tailored for AI workloads. Key features of the Summit include 200 Petaflops of processing capability, 250 petabyte storage capacity and speed of 25 gigabytes per second between nodes. Summit employs multiple hardware and software approaches to address data transport, connectivity, and scalability. Summit’s compute nodes each contain dual IBM POWER9 CPUs, six NVIDIA Volta GPUs, over half a terabyte of coherent memory (high bandwidth memory + DDR4) addressable by all CPUs and GPUs, plus 1.6TB per node of non-volatile RAM that can be used as a burst buffer or as extended memory. Second generation NVLink allows CPUs and GPUs to share data up to 4X faster than x86-based systems. Dual-rail Mellanox EDR InfiniBand interconnects, used for both storage and inter process communications traffic, deliver 200 Gb/s bandwidth between nodes. AI Bridging Cloud Infrastructure (ABCI) supercomputer has been commissioned by the National Institute of Advanced Industrial Science and Technology (AIST) in Japan and is being integrated by Fujitsu. ABCI is aimed specifically to offer cloud access to compute and storage capacity for artificial intelligence and data analytics workloads.

Dept of CSE, JSSATE

2020-2021

Page 12

NATIONAL STRATEGY FOR AI

4.2 ABCI INNOVATION PLATFORM

Figure 4.1

4.2 ABCI ARCHITECTURE The ABCI system will be using the new “Volta” Tesla V100 GPU accelerators, which sport Tensor Core units that deliver 120 teraflops per chip for machine learning training and inference workloads. ABCI is aimed to deliver a machine with somewhere between 130 petaflops and 200 petaflops of AI processing power, which means half precision and single precision for the most part, with a power usage effectiveness (PUE) of somewhere under 1.1, which is a ratio of the energy consumed for the data center compared to the compute complex that does actual work. The system is expected to have about 20 PB of parallel file storage and, with the compute, storage, and switching combined, burn under 3 megawatts of juice. The ABCI system will be comprised of 1,088 of Fujitsu’s Primergy CX2570 server nodes, which are half-width server sleds that slide into the Primergy CX400 2U chassis. Each sled can accommodate two Intel “Skylake” Xeon SP processors, and in this case AIST is using a Xeon SP Gold variant, presumably with a large (but not extreme) number of cores. Each node is equipped with four of the Volta SMX2 GPU accelerators, so the entire machine has 2,176 CPU sockets and 4,352 GPU sockets. The use of the SXM2 variants of the Volta GPU accelerators requires liquid cooling because they run a little hotter, but the system has an air-cooled option for the Volta accelerators that hook into the system over the PCI-Express bus. The off-the-shelf models of the CX2570 server sleds also support the lower-grade Silver and Bronze Xeon SP processors as well as the high-end Platinum chips, so AIST is going in the middle of the road. There are Intel DC 4600 flash SSDs for local storage on the machine. It is not clear Dept of CSE, JSSATE

2020-2021

Page 13

NATIONAL STRATEGY FOR AI who won the deal for the GPFS file system for this machine, and if it came in at 20 PB as expected.

Figure 4.2 ABCI Platform As per Fujitsu, the resulting ABCI system will have 37 petaflops of aggregate peak double precision floating point oomph, and will be rated at 550 petaflops, and 525 petaflops off that comes from using the 16-bit Tensor Core units that were created explicitly to speed up machine learning workloads. That is a lot more deep learning performance than was planned, obviously. AIST has raised USD172 million to fund the prototype and full ABCI machines as well as build the new datacenter that will house this system. About USD10 million of that funding is for the datacenter. The initial datacenter setup has a maximum power draw of 3.25 megawatts, and it has 3.2 megawatts of cooling capacity, of which 3 megawatts come from a free cooling tower assembly and another 200 kilowatts comes from a chilling unit. The datacenter has a single concrete slab floor, which is cheap and easy, and will start out with 90 racks of capacity – that’s 18 for storage and 72 for compute – with room for expansion.

Dept of CSE, JSSATE

2020-2021

Page 14

NATIONAL STRATEGY FOR AI One of the key features of the ABCI design is the rack-level cooling, which includes 50 kilowatts of liquid cooling and 10 kilowatts of air cooling. The liquid cooling system uses 32 degree Celsius water and 35 degree Celsius air. The water cooling system has water blocks on the CPUs and GPUs and probably the main memory, and there is hot aisle capping to contain it and more efficiently remove its heat. The HDFS file system that underlays Hadoop data analytics is a key component of the stack, as are a number of relational and NoSQL data stores. And while there is MPI for memory sharing and the usual OpenACC, OpenMP, OpenCL, and CUDA for various parallel programming techniques, and some familiar programming languages and math libraries, the machine learning, deep learning, and graph frameworks running atop the ABCI system make it different, and also drive a different network topology from the fat trees used in HPC simulations where all nodes sometimes have to talk to all other nodes.

Dept of CSE, JSSATE

2020-2021

Page 15

NATIONAL STRATEGY FOR AI

Chapter 5

RESEARCH AND IDEAS 5.1 HEALTH CARE The Government of India, through its recent policy interventions, has shown a bold commitment to achieve Universal Health Coverage and increased access to comprehensive primary health care. Through the Ayushman Bharat programme announced in Union Budget 2018, probably the world’s largest government funded health care programme, the Government of India has embarked on a path breaking journey to ensure the affordability and accessibility of healthcare in India. The Ayushman Bharat – National Health Protection Mission (AB – NHPM) aims to provide insurance cover of INR 5 lakh per family per year for secondary and tertiary care hospitalisation. Ayushman Bharat is targeted at more than 10 crore families (approximately 50 crore beneficiaries / ~40% of India’s population) belonging to the poor and vulnerable sections based on the SECC database, and doesn’t impose any limitations on family size or age limit for the beneficiaries to avail benefits. The benefits package covers most medical and surgical conditions with minimal exclusions, covers pre and post hospitalisation expenses, and covers all preexisting conditions from day one – thus simplifying availing requisite healthcare by the beneficiaries. The benefits of the Mission will be available at public hospitals as well as empaneled private health care facilities.

Figure 5.1 AI and Robotics in health The Union Budget 2018 also included a commitment of ~INR1,200 crore for Health and Wellness Centres (HWC), which will lay the foundation for India’s health system as envisioned in the National Health Policy 2017. These HWCs, to be set up by transforming 1.5 lakh Health Sub Centres from 2018 to 2022, are aimed at shifting primary healthcare Dept of CSE, JSSATE

2020-2021

Page 16

NATIONAL STRATEGY FOR AI from selective (reproductive and child health / few infectious diseases) to comprehensive (including screening and management of NCDs; screening and basic management of mental health ailments; care for common ophthalmic and ENT problems; basic dental health care; geriatric and palliative health care, and trauma care and emergency care). NCDs account for ~60% of mortality in India, 55% of which is premature. NCDs are predominantly chronic conditions and impact the poor most adversely, given the high costs of treatment involved. Prevention and early detection are therefore of the essence in reducing the disease burden attributable to these conditions as well as ensuring long-term follow-up and management of symptoms for patients. The HWCs, under the new implementation plan, will provide 12 basic healthcare services, expanding from the current package of 6 services. Crucially, these centres will provide preventive services to improve healthy behaviours for family health and control the incidence of communicable and noncommunicable diseases among the population covered by HWCs. A key component of HWCs will be universal screening for NCDs. Screening for five NCDs and associated risk factors has been prioritised given the high burden of disease associated with them. These include hypertension, diabetes, as well as three common cancers - oral, breast and cervical. Screening for other conditions such as Chronic Obstructive Disease will be added subsequently. The HWCs will be operated by a mid-level health service provider, auxiliary nurse midwives, accredited social health activists and a male health worker responsible for comprehensive primary health care services for a population of about 5,000.

5.2 AGRICULTURE AI will have significant global impact on agricultural productivity at all levels of the value chain. An estimate by Markets and Markets Research valued AI in agriculture to be USD432 million in 2016 and expects it to grow at the rate of 22.5% CAGR to be valued at USD2.6 billion by 2025. According to CB Insights, agricultural tech startups have raised over USD800million in the last 5 years. Deals for startups using robotics and machine learning to solve problems in agriculture started gaining momentum in 2014, in line with the rising interest in AI across multiple industries like healthcare, finance, and commerce. From analysing millions of satellite images to finding healthy strains of plant microbiome, these startups have raised over USD500 million to bring AI and robotics to agriculture. Globally, digital and AI technologies are helping solve pressing issues across the agriculture value chain.

Dept of CSE, JSSATE

2020-2021

Page 17

NATIONAL STRATEGY FOR AI

Figure 5.2 Accenture Research in Agriculture The relative role of each technology in creating impact is dependent on the nature of the work, and the issues at hand. India has ~30 million farmers who own smartphones, which is expected to grow 3 times by 2020 and 315 million rural Indians will be using internet by 202017. An Accenture study says – digital farming and connected farm services can impact 70 million Indian farmers in 2020, adding USD9 billion to farmer incomes. These are not futuristic scenarios, they are in play today, enabled by a vast digital ecosystem which includes traditional Original Equipment Manufacturers (OEM), software and services companies, cloud providers, open source platforms, start-ups, R&D institutions and others. Future growth is interdependent on the close partnership among these players.

5.2.1 AI SOWING APP Microsoft in collaboration with ICRISAT, developed an AI Sowing App powered by Microsoft Cortana Intelligence Suite including Machine Learning and Power BI. The app sends sowing advisories to participating farmers on the optimal date to sow. The best part – the farmers don’t need to install any sensors in their fields or incur any capital expenditure. All they needed was a feature phone capable of receiving text messages. The advisories contained essential information including the optimal sowing date, soil test based fertilizer application, farm yard manure application, seed treatment, optimum sowing depth, and more. In tandem with the app, a personalised village advisory dashboard provided important, insights into soil health, recommended fertilizer, and seven-day weather forecasts. In 2017, the program was expanded to touch more than 3,000 farmers across the states of Andhra Pradesh and Karnataka during the Kharif crop cycle (rainy Dept of CSE, JSSATE

2020-2021

Page 18

NATIONAL STRATEGY FOR AI season) for a host of crops including groundnut, ragi, maize, rice and cotton, among others. The increase in yield ranged from 10% to 30% across crops.

5.3 SMART CITIES AND INFRASTRUCTURE India is currently in the midst of a surge of urbanisation. While the percentage of the population living in urban areas was estimated to be 31% in 201122, recent research on satellite data indicates that this figure is close 45% today23, and predicted to rise to up-to 60 percent by 205024. Though seen as an important aspect of a country’s economic growth and a major step in the overall development of the country, unplanned urbanisation presents challenges such as congestion, over pollution, high crime rates, poor living standards, and can potentially put a huge burden on the infrastructure and administrative needs of existing Indian cities. To tackle these challenges, the Government of India has embarked on an ambitious initiative to set up Smart Cities across India, aimed at driving economic growth and improving the quality of life, by harnessing IT solutions. As part of the Smart Cities Mission, 99 cities have been selected with expected investment of INR2.04 lakh crores. The strategic components of these Smart Cities include city improvement (retrofitting), city renewal (redevelopment) and city extension (greenfield development) in addition to a pancity initiative in which smart solutions are applied covering large parts of the city. The Atal Mission for Rejuvenation and Urban Transformation (AMRUT) is another related initiative which targets improving the infrastructure of existing cities. Some use cases of AI that can augment the features of a smart city are listed below. a) Smart Parks and public facilities: Public facilities such as parks and other spaces contribute substantially to a city’s liveability. Use of AI to monitor patronage and accordingly control associated systems such as pavement lighting, park maintenance and other operational conditions could lead to cost savings while also improving safety and accessibility. b) Smart Homes: Smart homes concept is creating buzz with AI technologies being developed to optimise human effort in performing daily activities. Extending this concept to other domestic applications such as smart rooftops, water saving applications optimising domestic water utilisation for different human activities etc. c) AI driven service delivery: Implementation of AI to leverage data on service delivery could see application such as predictive service delivery on the basis of citizen data,

Dept of CSE, JSSATE

2020-2021

Page 19

NATIONAL STRATEGY FOR AI rationalisation of administrative personnel on the basis of predicted service demand and migration trend analysis, and AI based grievance redressal through chat-bots. d) Crowd management: Use of AI in providing effective solutions in crowd management in recent times have been in vogue and given fruitful results in averting city-scale challenges such as managing mega footfall events, emergency and disasters. Accenture worked with the Singapore Government during their SG50 Celebrations (50th anniversary of Singapore’ independence), and developed solution aimed at predicting crowd behavior and potential responses to incidents. The solution resulted in 85% accuracy in high crowd activity, crowd size estimation and object detection. Closer home, the “Kumbh Mela Experiment” is aimed at predicting crowd behavior and possibility of a stampede. Similar Big Data and AI solutions could help with advance prediction and response management. e) Intelligent safety systems: AI technology could provide safety through smart command centres with sophisticated surveillance systems that could keep checks on people’s movement, potential crime incidents, and general security of the residents. Social media intelligence platforms can provide aid to public safety by gathering information from social media and predicting potential activities that could disrupt public peace. In the city of Surat, the crime rate has declined by 27% after the implementation of AI powered safety systems.

5.4 EDUCATION In India, the importance of a developed education sector is amplified by a large youth population. Estimates indicate that currently over half the population of the country is below the age of 25. As the adoption of digital means of gathering data increases, it is important that these methods are effectively leveraged to deliver improved education and teaching. The adoption of technology in education is improving, though not at the pace required. It is estimated that schools globally spent nearly USD160 billion on education technology, or ‘EdTech’, in 2016, and forecast spending to grow 17% annually through 2020. Private investment in educational technology, broadly defined as the use of computers or other technology to enhance teaching, grew 32% annually from 2011 through 2015, rising to USD4.5 billion globally. Adoption of new technologies is still lacking, however, often attributed to unwillingness of teachers and students to adopt technology. School education in India has seen substantial progress in recent decades, with efforts at both the Central and State levels, and substantive gains in enrolment have been achieved – Gross Enrolment Ratio (GER) is 97% at elementary level and 80% at secondary level, as per recent figures. However, low retention rates and poor learning outcomes mar the impact of gains in enrolment Dept of CSE, JSSATE

2020-2021

Page 20

NATIONAL STRATEGY FOR AI

5.4.1 SMART CONTENT Content Technologies Inc. (CTI), an AI research and development company, develops AI that creates customised educational content. Using deep learning to absorb and analyse existing course materials, textbooks, and course curriculum, the technology creates custom learning materials, including textbooks, chapter summaries, and multiple-choice tests. A recent hackathon conducted by NITI Aayog also featured ‘ReadEx’, an android application that does real-time question generation using NLP, content recommendations, and flashcard creation.

5.4.2 WRITE TO LEARN BY PEARSON Pearson’s WriteToLearn software uses natural language processing technology to give students personalised feedback, hints, and tips to improve their writing skills. In describing his experience using WriteToLearn, one 7th-grade English language arts teacher said, “I feel it’s pretty accurate. … Is it perfect? No. But when I reach that 67 th essay, I’m not [really] accurate, either. As a team, [WriteToLearn and I] are pretty good.” Essay grading technology cannot substitute for a teacher’s ability to provide feedback and coaching on particular words and sentences: the software merely rates students’ essays in general areas— such as organisation, idea development, and style—and then provides generic suggestions for improvement in these areas. But when teachers use the software as a first pass at grading and then interject their detailed feedback to address the improvement areas identified by the software, essay grading becomes a much less time-consuming and laborious process.

5.4.3 PREDICTING DROPOUTS IN ANDRA PRADESH The AP government is making concerted efforts to bring down the school dropout rate in the state. It has tied up with Microsoft to address this complex challenge. Based on specific parameters, such as gender, socio-economic demographics, academic performance, school infrastructure and teacher skills, an application powered by Azure Machine Learning processes the data pertaining to all students to find predictive patterns. With these data insights, the district education officials can intervene and help students who are most likely to drop out. A variety of programs and counselling sessions could be conducted for these students and their parents. The Andhra Pradesh government, based on machine learning and analytics, has identified about 19,500 probable dropouts from government schools in Visakhapatnam district for the next academic year (2018- 19).

Dept of CSE, JSSATE

2020-2021

Page 21

NATIONAL STRATEGY FOR AI

Chapter 6

CONCLUSION Artificial Intelligence (AI) is bringing a dramatic shift in the world of technology where it can be applied for more productivity and success in order to simplify the system. From just your cell phone to the diagnosis of diseases, AI is now being used in many fields, offering high-performance and precise device operation with quality. In every area and not just technology, it has proven to be a path-breaking technology. As the fastest growing economy with the world’s second-largest population, India has a big stake in the AI revolution. The leading technology institutions in the country, such as IITs, NITs, and IIITs, have the ability to be the cradle of AI researchers and start-ups. In order to solve social problems, Indian start-ups are growing and developing AI solutions in education, health, financial services, and other fields.

Dept of CSE, JSSATE

2020-2021

Page 22

NATIONAL STRATEGY FOR AI

REFERENCES [1] Kathuria, Rajat; Kedia, Mansi; Kapilavai, Sashank. 2020. Implications of AI on the Indian Economy. © Indian Council for Research on International Economic Relations. http://hdl.handle.net/11540/12242. [2] Niti Aayog 2020. AIRAWAT-Establishing AI Specific Cloud Computing Infrastructure for India. https://niti.gov.in/sites/default/files/202001/AIRAWAT_Approach_Paper [3] Sunil Kumar Srivastava 2018. Artificial Intelligence: way forward for India. https://www.researchgate.net/publication/331972425_Artificial_Intelligence_Way_Forwa rd_for_India [4] Niti Aayog 2019. National Strategy for AI #AIFORALL. http://niti.gov.in/sites/default/files/ 2019-01/NationalStrategy-for-AI-DiscussionPaper.pdf [5] https://en.wikipedia.org/wiki/Artificial_intelligence [6] https://indiaai.gov.in/

Dept of CSE, JSSATE

2020-2021

Page 23