AI Deep Learning from -1 or 0 to 1: Lean-Startup Innovation as MVPs 3/3

Above picture is another overview how traditional analytics and newer “deep learning” techniques to consitute AI (Artificial Intelligence) in 2018 from the research of McKinsey & Company. Three most-wanted technologies are highlighted to explain or develop AI applications nowadays as advanced techniques: deep learning neural networks, reinforcement learning and transfer learning.

It can be a scaled-down option more focused on startup business' and/or technologies' modularisation, to continue the discussion in a profound way!

AI development as business-oriented way

Theory and practice differ from each other in many ways by their nature, but reflecting the same truth viewed just from different sides. The business version of AI developmentf needs strategically a new thinking at first, because a students' startup should not be as smaller versions of SMEs or large corporations. ?

Peter Thiel? - the author of book "Zero to One" is one of Silicon Valley's most successful tech entrepreneurs and investors. A good book is able to explore a topic in a deeper way and help those learning by doing as the readers. What can be found theoretically from his book if AI Deep Learning also from -1 or 0 to 1?

He points out 7 questions that every startup should answer:

1. The Engineering Question: Can you create breakthrough technology instead of incremental improvements?

2. The Timing Question: Is now the right time to start your particular business?

3. The Monopoly Question Are you starting with a big share of a small market?

4. The People Question: Do you have the right team?

5. The Distribution Question: Do you have a way to not just create but deliver your product?

6. The Durability Question: Will your market position be defensible 10 and 20 years into the future?

7. The Secret Question: Have you identified a unique opportunity that others don’t see?

To avoid disposing of confidential information, the most of answers are not possible to give in details. However, the last 2 questions are worth to look over openly for further discussing, so as to describe the hungry for victory in a long-term view, as well as strong sense of killer instinct. Those are what for the startups must to have in surviving and/or growing better.

Sky-high ambitions: defensible in long-term?

To a startup, the preoccupation is concerned much healthier not only with short-term earnings - as long as anything brought in cash - but also to its long-term ambition doable under such cash-flow.

Scandicode's cognitive assessment solution will be explored with those advanced technologies in Header Image, apparently continuing "Pre-MedTech AI" prototyp. It makes sense for new modular parts to approximate human cognition in AI design, more exactly as "Deep Reinforcement Learning". It stands for Scandicode's AI interests to nail down as follows:

Ambition 1:

Here are some details as must-take challenges in long-term if to simulate the neurons interacting in human brain:

• The real-world and synthetic datasets across cognitive domains can exercise IT or supervised ML (Machine Learning) functions, but to give more autonomy for deep learning neural networks, all performing interactively with less human intervention in external validation-loop as the goal.
• The reinforcement is about such an ideal system (similar to AlphaGO & AlphaGO Zero as 2in1) receiving virtual “rewards” or “punishments” added on unsupervised-AI as an internal verification instead, essentially learning by trial and error over the execution before the outcome having enough multi-evidences. 

Ambition 2:

AI to health technologies can be thought not as one-off experimental project due to lack of development tools (in comparison with projects in Business Intelligence field). The ambition is much more defensible as an evolution to gain continuous intelligence, like a big diamond willing to get cut. It will be wiser to learn about the advances continuously over time from AI applying to image generation, style transfer, or even medical brain CT images. 

For instance in those 7 alternative use cases, such synthesis of generative models can be seen or already discussed in different areas, manifold hypothesis, and disentangled representation:

1. Data augmentation; 2. Privacy preserving; 3. Anomaly detection; 4. Discriminative modeling; 5. Domain adaptation; 6. Data manipulation; 7. Adversarial training

Ambition 3:

Of course, it will be so expensive to handle MedTech AI with regulatory compliance. A company has to need huge effors and deep pockets whenever staying in the game, so that the answer to our secrete question becomes very unique - especially to emphasize transfer or multi-task learning:

(The picture from: https://www.frontiersin.org/articles/10.3389/fphar.2018.00074/full)

Transfer learning: urgent or more suitable from -1 or 0 to 1?

AI is rapidly in changing at breakneck speed. Its future applications will be optimised with colossal training datasets or computing time. It may still be the result of many years for Scandicode Oy's special MedTech AI to excel in new innovations, although elderly wellbeing sector or eldercare industry are now targeted as an acceleration. Transfer or multi-task learning can get much of the benefit to survive the valley of death as the goal (reasonablly not limited as a secrete) for startup business:

• In particular, transfer learning has been newly exciting to reuse prior work or a pre-trained model on a new problem, transfer domain expertise from one task to another, democratize machine learning models and so on, similar as the modularisation literally for AI much more explanable and understandable to business world.
• Gartner research indicates "85% of CIOs will be piloting artificial intelligence programs through a combination of buy, build, and outsource efforts" by 2020. For MedTech and/or any industry that truly need AI soon for strategic experimentation, it can be necessary to hire more experts and/or purchase specialized services.
• Effectively without making enormous investments, Scandicode Oy is able to turn its experience so far for such exploration as a paid service, which is where transfer or multi-task learning comes in. Same AI can be applied to other fields without losing the focus of knowledge growth or own engineering talents in teamwork, such as Business Intelligence or Intelligent Automation also suitable to the collaboration.

Thus, the emphasis to transfer or multi-task learning can be as the key not only among medical or health related domains for technical development itself, but also between AI furtherance and startup business, to solve new challenges more quickly in real world.

Conclusion

OBSERVATION 1

• Scandicode Oy should be able not only to help its 80 - 90% of personnel back to normal SMEs so far or further TE training at any time, but also better to go a step further for creating new service business, as long as it brought in cash to support own employment.
• Either way, the contribution benefitial to local society is a common goal once achieved for solving the pain or creating a gain!
• It will be much more proud as own gowth success via new product or service if AI technological advances can reshape healthcare, elderly wellbeing, or eldercare sectors, as well as any industry / company requiring such innovative value.

OBSERVATION 2

• To ex-Nokians or other ICT professionals in uneployment status, MedTech or health AI has been proved as an attractive direction to update good employability preferably via supporting students's startup as a win-win approach.
• With transfer learning to and/or into AI (such as cognitive science, geriatrics, healthcare or health, medical device, regulatory compliance, ICT, BI, sales, etc.), it can enjoy more in immediate scheduled tasks for raising or saving ourselves as middle- or older-aged adults mostly under mortgage stress, especially in age group of 55-64 years old. Learning mind is a choice to allow our brain with more links between neurons to strengthen in challenges.
• Albert Einstein said: “It is not that I’m so smart. But I stay with the questions much longer.” “Learning is experience. Everything else is just information.”A painful learning experience about unemployment or startup adventure will make us in AI adopting and developing too happy to grow again.

OBSERVATION 3

• The customers from a wider range of industries can have business-oriented approach now for their AI exploration experimented in comparison with Scandicode Oy's AI prototype or own project.
• At least, Scandicode Oy's service will be a cost efficient way to build Proof-of-Concept from scratch and understand its social impact such as CE regulatory compliance. 
• It can be as an assignment with tremendous discount if the scope is belong to "Deep Reinforcement Learning" same to Scandicode's AI interests. With a little healthier cash-flow, it will get better chance to own products survising or contributing more in the fields for elderly independent living and successful aging. 

As relevant technologies innovating so rapidly, AI (Artificial Intelligence) development in startups is very much alike to jumping off the cliff and building own wings on the way down. Along with the death-valley curve, it should be so important always to sharpen different or new advantages / not to yield and gamble with the disadvantages. A young startup has to be thus focused more on developing the opportunites as its way to solve the problems, not just limited by the challenges but challenging the limits instead.

Athough the second post is clearly focused on more technical details, other 2 posts are more open to a business-oriented way beyond the technology itself. All those can reflect a journey how young startup to keep its growing from "visions or destinations (as WHY in Panic Zone) vs. goals or deliverables (as WHAT in Stretch Zone) vs. tasks or milestones (as HOW in Comfort Zone)" till adequate to "flying" similar with a business GPS - targeted on AI technology implementation increasingly via learning by doing.

Due to my background from IE (Industrial Engineering), case study will continue AI Deep Learning from -1 or 0 to 1 and proceed new know-how as a research. Scandicode Oy's story is on-going as startup wisdoms or lessons aren’t previously taught in class and textbook, happy to learning and catching much more in the future!

AI Deep Learning from -1 or 0 to 1: Lean-Startup Innovation as MVPs 1/3

AI Deep Learning from -1 or 0 to 1: Lean-Startup Innovation as MVPs 2/3

AI Deep Learning from -1 or 0 to 1: Lean-Startup Innovation as MVPs 3/3