Hi Smartonso I share your enthusiasm for Tata Consulting Services collaboration with BRN and agree that the full import has not been properly appreciated by the ASX.
In this regard though I have posted the complete paper published in December, 2019 by Dr. Arpan PalChief Scientist and Research Area, Head Embedded Systems and Robotics TCS Research Tata Consultancy Services India titled "Using Ai on Iot Sensor Data - for predicting health of man and machine" has either not been read or understand by other than a few and so I have extracted the following passage which clearly in the circumstances of the joint presentation at NeurIPS 2019 must be based upon his knowledge of and involvement with AKIDA:
4) Non-availability of data and labels –Availability of sufficient data to train AI models is always a problem which is more pronounced in deep learning based systems. Even if data is collected and models are trained for one scenario (may be one type of factory or health data from people a particular country), there is no guarantee that such a trained model will work in a different but similar scenario (another factory with similar machines or another country with different demography people). This raises some few very important but practical aspects – a. Few-shot learning[23] and Meta-learning [24] – Systems should be able to learn quickly on a few instances of training data and should be able to use meta-knowledge available to augment the data learning. b. Unsupervised learning and Transfer learning –Systems should be able to infer reasonably in the absence of labels or where labelling can be done on demand by human experts on a reduced subset of the data identified by unsupervised approach. Transfer learning techniques can help in re-training existing pre-trained models from one scenario data set with a small representative data from the new scenario. 5) AI at the edge – Edge devices / on-premise devices play a large role in IoT systems. In the context of AI based analytics, they play significant role to provide a. Low-latency,real-time inferencing needed for IoT-driven process control systems. b.Low-battery consumption that is needed for energy-constrained devices likewearable and implantable. c. Privacy-preserving analytics as the data does not leave the edge / premises even for analytics. AI at edge either needs special technique to compress the AI models enabling them to run on constrained edge devices, or have dedicated low-latency, low memory inferencing algorithms, or have special purpose hardware accelerators in the edge [25], [26]. In order to reduce the energy consumption by significant order, completely new processor architectures called neuromorphic [27], that mimic the brain in hardware, are been used to design new chips - such chips, coupled with new sensing techniques called spiking sensors and 3rd generation brain-inspired neural networks called spiking neural networks (SNN) [28] hold the promise of disrupting the low-power edge AI technology.
BRN Price at posting:
6.3¢ Sentiment: Hold Disclosure: Held
(20min delay)