Transer Learning : Super Intelligence

Transcription

1 Transer Learning : Super Intelligence

2 GIS Group Dr Narayan Panigrahi, MA Rajesh, Shibumon Alampatta, Rakesh K P of Centre for AI and Robotics, Defence Research and Development Organization, C V Raman Nagar, Beangaluru

3 Artificial Intelligence (1) Algorithms Enabled for Constraints Exposed By Representation that Supports Models Targeted At Thinking, Perception, Action

4 Artificial Intelligence (2) The Branch of Computing Science attempting to built machines and systems capable of intelligent behaviour

5 Artificial Intelligence (3) It is a discipline to impart ability into a machine to learn rather than programmed explicitly. (All possible data, events, behaviour, action) Machine Learning (ML) ML<AI

6 Genesis and History of AI Turing Test (1950) Summer School in Dartmouth College, Hannover, US (1956) by John McArthy the genesis of the name AI Winter of AI, All academic papers but no commercial exploitation/ no application/no system Small AI spring

7 Domain Specific Applications of AI Intelligent Search Engines On Line Portals Suggestions for Turbine Maintenance Medical data diagnostic (Cancer Detection) Credit Rating of Enterprises Customer Profiling for retail malls Face recognition Anomaly Detection

8 Artificial Intelligence Un- Supervised Learning Deep Learning Machine Learning Supervised Learning Transfer Learning Artificial Neural Network Convoluted Neural Network Super Artificial Intelligence

9 Contributing Fields of AI

10 Human Learning System

11 Human Intelligence System

12 3D-Model of a Neuron

13 ANN(Modelling Human Learning)

14 Domain Specific NN (Convoluted NN)

15 Deep Learning through ANN

16 Machine Learning through NN Domain Specific Learning of the ANN Through Supervised Learning Technique Multi Layer Perception Network With Feed Back Tunnelling for Efficient and high speed Convergence Re-enforcement Learning from marginal attributes

17 Transfer Learning Transferring the Learning that has already Learnt through deep learning. Input is about time series events (Time-series data) Through on line sensors. Earth quake data, Seismic data, Patient vital records, Weather data, Stock market data, Social network data...

18 Convoluted ANN

19 Machine Learning ML>NN>Convolution NN>Deep Learning>Transfer Learning ML (Supervised Learning With Feed back) ML(Unsupervised Learning (Without feedback... Search for things we have no clue about) No Domain Specific, unstructured data, NLP

20 Applications Anomaly Detection (Gesture Learning) for Disaster Prediction, Identification and mitigation Speech Synthesis/ Speaker Identification / Speaker Recognition Motion Picture Based Lost Human / Child Tracking. Assert Monitoring Smart Parking, Resource Utilization, Smart City

21 Cyber Defence Monitoring IP Network with Cellular Phone Network and IoT, has created an unprecedented NW-of-NW It Has increased the profile and attack surface of the network First detection of anomalous behaviour of such network is beyond action of few professionals. Cyber Traffic Profiling, Clustering, Anomaly Detection, Fishing detection, Counter attack Planning AI based cyber traffic profiling and monitoring is only solution.

22 AI For Cyber Security

23 MIT builds AI based system that can detect 85% of Cyber Attacks AI2 : Training a big data machine to defend Veeramachaneni, Kalyan, et al. "AI^ 2: Training a Big Data Machine to Defend." Big Data Security on Cloud (BigDataSecurity), IEEE International Conference on High Performance and Smart Computing (HPSC), and IEEE International Conference on Intelligent Data and Security (IDS), 2016 IEEE 2nd International Conference on. IEEE, 2016.

24 Robotic Applications Autonomous Navigation System / Mobile Robotics Platform(Land, Air and Water) Simultaneous Localization and Mapping (SLAM), Vision, LiDAR and RGB-Depth based Real Time Object Recognition and Tracking Human Activity Analysis

25 Autonomous Unmanned Systems are used in different warzones to enhance the operational capabilities of the military. UAV for Surveillance and Reconnaissance purpose AI technologies that is used to synthesize 3D world model of the environment, demonstrate perception informed autonomous navigation, automatic object classification and detection, Multiple level of situation awareness

26 PGM mounted on a Tracked Platform

27 Humanoid Engaging a Target

28

29 IA Applications Recognized as AI Applications Game Playing General Problem Solving Expert System Natural Language Processing Computer Vision Robotics (AI on Wheel)

30 Disruptive Innovation using AI AI is coming Technology is making it happen possible doable useful viable And is very powerful Your digital transformation through AI is inevitable Intelligent amplification of machine better than human, Automation of Intelligence are on the way

31 Road to Super Intelligence Artificial Super Intelligence (ASI) Artificial General Intelligence (AGI) Artificial Narrow Intelligence (ANI) (Dr Watson)

32 Advertisement Job Code: 011/2027/Jan 15/220 Job Profile: To Search 10 TB of data and detect anomalous behaviour and report to authorities within 20 minutes Who can Apply: Human with domain specific knowledge / Machines capable of TB data search and anomaly detection Only machine with having capability of scanning 20TB data and predict anomaly or event likely to happen 20 minutes in advance can apply

33 Thank You

34 Applications of AI more of IA

35 Spatio-temporal Object Detection and Classification This isolated road network is an object, one of whose behaviors is that it must be connected... to the main road network. So the operator will be prompted during production, densification or validation A behavior of a major road is that when it intersects a river, there must be a bridge. Therefore the operator will be prompted during production, densification or validation. This lake object has behavior that makes it display as a complex polygon.