Back

Visual AI for a more Human AI


 Publish Date :2019/07/17

The final session of the AI forum is a panel discussion featuring panelists Dr. Albert Liu, CEO of Kneron and Mr. Vitaliy Goncharuk, CEO of Augmented Pixels. The panel was moderated by Mr. Limas Lin, Vice President of Etron.

Point Cloud and Edge AI Net

SLAM (Simultaneous Localization and Mapping) is the core technology for robotics, AR/VR, etc. The technology from Mr. Goncharuk’s company, Augmented Pixels aim to create maps of all the entities in the environment and then have the entities run simultaneously and interact with each other. SLAM requires both hardware & software and partnering with companies specializing in visual sensor hardware such as Etron is greatly beneficial for Augmented Pixels. This is true especially because their technology focuses greatly on Point Cloud, the same technology used in AR and VR glasses to see and understand the environment.

Mr. Goncharuk stated that most consumers care only about the final results and how the robots can solve their problems; the specific details and what happens under the hood is not their concern. To achieve the optimum final solutions, all the related components must be optimized as well.

Dr. Liu aims to build up a more flexible AI chip which can enable almost all the devices. At the same time, he is also trying to build up Edge AI net to connect humans to device & device to device and is able to define the function based on the device with solutions inside. Edge AI Net should be able to define all the devices with different functions and can be changed from time to time.

To achieve this, they made a reconfigurable AI chip that can enable AI features in wearables and smart devices. Because the solutions and platforms are the same, the devices can communicate with each other. The best advantage of this reconfigurable AI chip is because it can be used for multiple applications; 2D or 3D sensing, voice based command, and more; depending on the requirements of the AI model. Operating on the edge will provide better privacy, more real-time operation, and cheaper operation cost.

Algorithm, Dataset, and Accuracy

Mr. Lin shared his experience of talking to Professor Andrew Ng of Stanford University where Professor Ng shared the 3 key metrics of AI: the quality of their algorithm, the amount of datasets available, and the accuracy of the final results. As such, these 3 key metrics can be used as benchmarks of the quality of the AI.

Dr. Liu stated that customers of many AI companies are internet companies and cooperating with them can yield significant datasets. Having an international presence will also help companies improve their positions in the metrics as well as the company can then capitalize on the strengths of specific regions. In Kneron’s case, they obtain their datasets from China, software development is done in US, and their hardware is sourced from Taiwan.

Mr. Goncharuk offered an alternative where synthetic datasets are created to train the AI models. Startups and many smaller companies cannot compete with the larger companies due to their inability to secure sufficient amount of data. Using the synthetic datasets might be able to put all the players in a more equal footing. For algorithms, the main problem is the implementation aspect; how the company can implement AI solutions into actual business processes and bring value to their customers. In his opinion, the main challenge for the industry is how the technologies will be used and implemented in the future or even, if a general AI will have been created by that time.

Preventing Evil and Evil AI with AI

As with other technologies, there are always potentials of AI being utilized for nefarious purposes. In Mr. Goncharuk’s opinion, there are no differences between bad people & preventing their behavior with bad AIs and preventing their behavior. As AI becomes more common and more intelligent, there will be a big industry in preventing bad AI and protecting people from bad AI. He hopes that the industry will balance itself between the bad and good actors to prevent a possible catastrophe.

By contrast, Dr. Liu opined that most evil AIs come from the centralized one; which is why his company focuses on edge AI. One of the biggest visions in edge AI is to distribute the power of AI so that it can be self-learning in the edge AI to be more human: individuals and diverse instead of a hive mind.

A Sense of Purpose in AI

AI is expected to achieve greater things that were inconceivable just years ago; however the limits of its abilities or what it can achieve are still unknown. Mr. Lin’s purpose in AI was stated in his keynote speech; which is to educate the young and care for the old.

Dr. Liu put a higher emphasis in AI due to a few tragedies in Taiwan that happened several years ago; that might have been preventable with AI. In one of the incidents, Dr. Liu stated that cameras were able to track the perpetrator, but no one could do anything nor could the cameras send any form of warnings. His idea then was to bring AI features and functions to devices; enabling communication between devices when they notice abnormal behavior. He believes this solution can improve lives; making it safer and more convenient.

Mr. Goncharuk was of the opinion that AI is an inevitable future. The competition in AI is fierce and the countries and companies doing AI, computer vision, and deep learning are expected to be more powerful and influential. Joining this competition will require a deep understanding of the current trends and situations of the industry. The competition is still ongoing and from it, many interesting solutions have been created.

Working Together to Enhance AI

Dr. Liu stated that in any industry, the only plausible way to create true success or improvements is to define a solution to fix the customers’ problems or to make their lives more convenient and this is an aspect that AI excels in. Currently there are many types of sensors, but they are not yet able to filter what they are detecting or do selective focus. A brain or engine in the sensors will be needed to analyze and recognize a particular abnormality, making vision for AI more pragmatic.

According to Mr. Goncharuk, vision for AI depends on the quality of the cameras and to enhance AI, they can share best practices with each other and establish standards to follow. He believes this will allow companies to deliver good solutions for the customers. AI developers also need to reach the first layer of abstraction as soon as possible as the customers will need it also as soon as possible. The created solutions need to be simplified so and the focus will need to shift from the technology to customer experience and solution simplification.

Dr. Lin agrees with the panelists’ opinion. He shared that when working with customers for the AR/ VR solutions, a heavier emphasis is put on the user experience, not the cost per component. He stated that to achieve the best imaginable user experience, cooperation between companies will be necessary.

To watch the full forum session, visit our YouTube channel here.

Back