CHAMP- Concept Hierarchy Aggregator, Mapper and Projector

Learning how to self-learn is difficult. Most self-learners don’t know where to start, what to cover and what concepts relate to each other, and how. Usually, it's only truly acquired during Ph.D.

CHAMP is a NLP concept hierarchy module capable of analyzing every skill, breaking it to sub-skills, guide a user how to climb from skill to skill, assessing user's proficiency in a given skill, asses the quality of the user's work using a given skill and its topic coverage, as well as its match to intended audience.

It can check if work is of progress over existing knowledge, and refer it to experts should the writer want it, detecting talent early regardless of location, gender or class.  

It can examine the learning plans made by ministries, and give the minister a cover in topics the minister has no expertise.

This is CHAMP, the future of training. 

The problem I'm trying to solve is making self-learning an ordered process rather than controlled trial-and-error chaos.

The problem is innate to the process of organizing knowledge, and can be avoided using learning frames such as universities, which have experts that arrange the material and offer guidance should one fail to understand anything.  

However, many cannot afford university, and even those who can leave it at some point and must fend for themselves in the free market. There, most fail to acquire new skills as they don’t know how to arrange them, and lose their job when the market changes.  

There probably isn’t one human being that isn’t affected by it, but my focus is on under-privileged populations, such as those who can't afford education (moneywise or time-wise, the effect over their lives is the same), as they benefit most from it and are most likely to adopt it.

Both in Latin America and the Caribbean, there are traditionally under- privileged populations, with no fault of their own. These people usually have little resources available, and cannot propel themselves higher as a result of that. Societal prejudice, which is also present, isn’t helping their situation either.

CHAMP can change that. 

I serve any human being wishing to gain or improve a skill demanding theoretical knowledge.

The solution addresses their needs by mapping for them what they need to do to get to their goal, from their current point of proficiency, throughout time and for free.

They will no longer need to read material to understand if it deals with the skill they wish to learn- CHAMP can classify it for them, letting them focus on what actually matters- learning what THEY need to know to acquire skill, without missing or repeating material.  

The nature of the problem is no different going up the educational and socioeconomic ladders, but the impact is- a person with computer sciences bachelor degree in the US not knowing what to train further to get ahead at work is facing the same problem as an underprivileged person not knowing how propel to better future in Haiti, and the solution can help them both. That being stated, the dreamer in Haiti is far more likely to use it, as the incentive is higher.  

My solution is a novel NLP software module, working in the following fashion:

Conceptual hierarchy mapping is the use of the links between concepts to understand their relation.

It does it by determining how many concepts are needed to be known in advance in order to define the current concept, and how many concepts rely on it, or in other words, how abstract and how fundamental a concept is.

As skills are concepts by definition, the method can be applied on them.

The method is relatively simple and effective, can automate the process of knowledge arrangement, and can analyze the novelty element in written work.

Abstraction is the ability to use concepts to define more complex concepts, enabling humans to deliver huge amounts of information in one sentence, or even one word.

For example, we use the concept dexterity to describe coordination of small muscles, and particularly fingers, with human eyes. Each of these concepts is also defined by other concepts, until raw sensual data (precept) is reached.

The amount of concept layers needed to define a concept is called “abstraction level”. The higher the abstraction level, the more complex the concept is and the harder it is to learn.

The more concepts rely on another concept, the more fundamental it is.

Complex skills are found by analyzing abstraction level, i.e. how many concepts are needed to define the examined skill.

Foundational skills are found by analyzing reliance level, i.e. how many other skills need to use the examined skill in order to be defined.

In CHAMP's current implementation, it is done by mapping concept prevalence in lexical dataset.

Adjusting for previously learned concepts is rather straightforward- if a concept has been learned, its abstraction level will set to be zero.

When a new piece of knowledge is introduced, it can be mapped in the same fashion and compared to existing concept mapping. Contradiction in the concept maps will indicate wrong data, and additions nonexistent in it will indicate new data.

A skill containing only zero-level abstraction concepts is considered learned.

By subtracting sub-concepts of previously-learned skills that overlap those of the desired skill, a map of the exact additional concepts one needs to learn to acquire the new skill is made.

This creates roadmap from any skill to another, similar in function to skills trees in computer games, but for real people and with real data.

My solution is innovative by addressing written data not by traditional NLP tools, but rather by addressing the hierarchy of the concepts in it.

When it comes to skills, it is particularly effective, as skill are basically concepts, and as such, they are composed of simpler mini-skills and compose more complex (or abstract) skills.

For example, the ability of a surgeon to perform surgery requires, among other skills, the skills of stitching, anatomy proficiency and dexterity. Without one of those, the doctor cannot solely perform surgery.

Other NLP tools are not yet capable of mapping concepts, and therefore cannot map knowledge, and among that skills. CHAMP is, and that is its main innovation.

Additionally, when implemented on skills, CHAMP maps skills and sub-skills and creates a real-world skill map, showing the user all the steps needed to gain a skill, and what is needed to learn to climb from skill to skill.

This is a significant leap from traditional curriculums, and this makes CHAMP useful in any stage of professional life and for any knowledge the user wishes to gain.

In the novelty analysis aspect, CHAMP is particularly useful in defining if something composes a concept, and therefore, if something does not. This makes it very useful in defining whether something is new, contradicts existing knowledge or adds on it, not by trying to make a machine "understand" the text but by mapping the concepts in it. This, again, is innovative.

My solution relies heavily on the foundations made in Ayn Rand's theory of concept formation.

A POC of the module can be found in this Google Colab Jupyter notebook:

https://colab.research.google.com/drive/1e5TXdVMvacftmGK9FW83yXQJN8NlqGgf

The number of people served now is zero (it was born several months ago).

The number of people served in one year is expected to be around 92500.  

This assumption is assuming early-adopter global ministry of education cooperation rates, 2.5%, and early-adoption rates, also 2.5%, or 0.0625% adoption rates, multiplied by the amount of people aged 10-24 in Latin America and the Caribbean, 148 million. 

The number of people served in five years, assuming cooperation with validation-providing establishment of reasonable tier (threshold- ranked in the Shanghai 500), is expected to be 16.8 million.

This assumption is based on course web-platform user data, which are 24 million Udemy users and 30 million Coursera users, assumes 50% user overlap between the platforms, and product-market-fit threshold adoption rates, 40%.

*. The reason for using different user bases in the one-year and five-year projections is that in the short term, the user base will come by cooperation with ministries of education, and will not grow organically.

In the long term, product-market-fit and organic growth will be what keeps users, and the closest appropriate examples of products and their user base are the course web-platforms. The user base itself might be different, but adoption rates are expected to behave the same, and hence that user base is expected to provide more accurate long-term projections for the solution user-base.   

Next year, I would like to have a minimal valuable product (Q2/Q3 2020), bring it to product market fit (Q4 2020), and find an academic validation partnering institution (Q3 2020).

Within five years, I would like the solution to be the standard of skill acquisition, meaning that any person wishing to learn a new skill will use CHAMP, or CHAMP-variant, to assess what is needed to learn, and employers will use CHAMP to assess potential workers as standard procedure.  

As CHAMP has concept hierarchy mapping capabilities, I would also like it to be a benchmark in novelty analysis, meaning that every person coming up with something new will display CHAMP analysis to prove that.

Figures-wise, a projection of around 17 million users, 1.2 million in revenue, and at least one Ivy-League university as a research and validation partner seems viable.   

Impact-wise, this changes everything. People without access to education will be able to learn on their own, employers will be able to assess that learning, researchers and inventors will be able to show innovation without expensive patent searches and literature reviews, and most importantly, anyone willing to use their brain will be credited for it.

Please let me be clear- I aim to change not only the Latin America and the Caribbean skill acquisition market, but the worldwide one.

I wish to change the way humans seek, select and improve skills, and I built a tool that has the potential to accomplish that if used correctly and applied in scale.   

Financially, in the beginning years, this project will incur costs and have little to no revenue. This is always problematic when raising funds, and particularly when raising funds for largely free-usage projects. 

Technically, in the implementation perspective, there were some issues with determination of the weight given to each sub-skill. This is important the deeper skill analysis goes.

In the scale perspective, some elements in CHAMP are not expected to scale well in their current configuration, and extra attention must be given to performance upon scale due to the nature of the module.

Culturally, I expect significant old-guard resistance to the solution- I seriously doubt that people who are used to deciding what will be learned, where, and by whom, and as a result, who will get hired and who is judged to spend life in the gutter, will be willing to give up that power so easily.

Additionally, I expect older generation resentment, at least in initial stages, towards the capability of a machine to arrange concepts, as automation of distinct human capabilities almost always generates fear.   

Market-wise, job market adoption of validation generated by the solution as valid, as well as using the solution for staff recruitment tests, is expected to be lengthy (years) as a result of conservatism in most job markets.   

Traction-wise, I cannot generate the traction such a project needs on my own. Even in my semi-academic hat, or via my research work, I am not exposed to enough people to really generate worldwide traction. 

Financially, I intend to use grants, prizes and other unstable income sources as runway-fueling elements, and generate revenue from employers and learner displayed ads and paid-tier to fund the rest of the operation.  

Technically, it seems that a new method I implemented has resolved the weighting issue, but more testing is needed to be certain, a process I expect to be done (assuming reasonable re-iteration cycles) in around 14 days.

In the scale perspective, extra attention will be given to performance upon scale, slowing development but enabling scaling.

Culturally, the old-guard resistance to the solution will be met by ignoring it and offering Ph. D. students and inventors the ability to check their work for novelty using CHAMP. Once they begin introducing it to advisors and venture capitals as validation data, it will be adopted by the advisors to check the novelty of their own work and by venture capitals as a pre-screening tool. Once that happens, the old-guard becomes irrelevant.

When it comes to older generation resentment, it must be endured as this is a natural reaction to change with little long-term effect on it.   

Market-wise, I will simply start with those who adjust fastest, the big-tech companies, and let the rest of the market follow.   

Traction-wise, I intend to partner with institutions that can generate worldwide traction, such as top-tier universities. I believe that will be easy once the module scales and they see potential. 

As part of my expansion plan (or should I say, deployment plan), I want to focus on a population that benefits greatly from CHAMP, and the skill gap in Latin America and the Caribbean is both a great test case and a golden opportunity.

I believe that there is a golden opportunity for early adoption of CHAMP among no-to-mid income women in Latin America, but I am not certain I am the best person to reach them. For that, I intend to locate a regional distribution partner. My ideal distribution partner will be able to reach the ministries of education, as they can grant CHAMP access to millions of benefiting users.

A university, for example, will match perfectly, but I am open-minded to any partner bringing value to the table.

The currently planned incentive for the distribution will be either revenue sharing or grating the rights for some or all employer usage fees in the region, depending on the result of negotiations and the impact of the partner.   

As the solution is a software module and as said software module can be implemented in other software products, adaptation will not be an issue. 

The solution team, being myself mostly for now, is supported by my dev-shop, IGPT Innovation, in terms of staff (5 strong) and code base, when needed.

I own the dev-shop, so I control it and can make such calls.

The dev-shop has no financial interest in the solution, as it makes its profits elsewhere, and all cooperation and contribution is done pro-bono.

Should I need to recruit additional people for the solution, my dev-shop staff will be my first pick, and as I control it, I can make whomever I need available for the specific periods of time I will need them for.

Should the solution be successful, publication for the dev-shop, and credit to its staff that had part in said success, can be a bonus, but isn’t a must.     

Actual Headcount

In-house staff: One part-time (myself).

Contractor: Dev-shop (5 strong), need-based.

Work Deviation Count

Due to its nature, selected structure, and work done so far, I estimate that the CHAMP requires innovation only in the concept formation modules, which is around 10% of the total work.

The rest, around 90%, is infrastructure and scale, and a robust code base for that part is available in my dev-shop, which is willing to donate a copy of the needed parts to the project. Adaptations will have to made, but it’s quite a jumpstart and saves quite some labor costs.

I and my team (and by team, I mean my dev-shop team) are the best-placed to deliver this solution because of the following:

I sold several pending-patents to Fortune 500 companies and led several development projects for Fortune 500 companies.

I am responsible for 5 breakthrough-level inventions.  

Products my dev-team and I developed are used in high profile bodies, such as the Israeli premier league, in my home market of Israel, and were covered on Israeli TV.

Academy-wise, I am the CTO of an under-establishment medical interaction research center in the Coleman College of Management in Israel, and I am a mentor in the entrepreneurship program of Herzelia Interdisciplinary Center (IDC), which is considered to be the best of its kind in Israel.

To conclude, I know small and big scales, I can innovate, I can research, and I come with a team of 5, all of high caliber and all available to me for the exact intervals I need their help in.

Also, I invented CHAMP and concept hierarchy mapping, so I'm probably the most qualified to work on its applications in this point in time.    

I do not partner yet with any organization.

The key users can be divided into three groups:

Learners

Any user wishing to learn something new or build a skill map is a learner.

They will be able to select any skill they want to acquire, as well as feed the ones they already have. CHAMP will then provide them with a stepping stones map for getting there.  

The service will be given for free.

As they advance in skills, they will be able to test their knowledge using CHAMP selected questions based on their knowledge.

The service will be given for free with ads or paid without ads.

Pioneers

Inventors and researchers will be able to analyze their work using CHAMP, showing whether or not it is novel.

The service will be given for free, and it is meant to make CHAMP a benchmark service in innovation.

Employers

Employers are the key customers of CHAMP, and will be offered worker skill validation feature.

CHAMP will read potential worker's CV, and assess which mentioned skills are relevant for the position. Then, CHAMP will prepare a set of test questions (using a CHAMP mined question bank) fitting the skills the user reported having, in random and in increasing hardness order.

Theoretically, every skill the user declares about can be tested this way cheaply. Skill test results can be saved in one job interview and presented in the next one, assuming both places use CHAMP.

Fees will be charged per examined user, with unlimited number tests per user. 

Zero stage of this solution was funded of my own pocket, with considerable donations in workforce and code from my dev-shop.

Investment capital is unlikely in a project of that kind, as the inclination of investors will be to make the technology proprietary and getting the revenue stream that way, a direction I prefer to avoid as I believe the direction presented below will be better.  

The first stage, i.e. the stage in which users generate little to no revenue, is expected to be funded by donations, grants and hopefully, sponsorships by an organization larger than mine (Google will be perfect, though unlikely), as well as by revenue from worker skill validation.

Realistically, financial stability is not expected until the serving stage.

The serving stage, i.e. the five-year mark goal (CHAMP is a worker assessment standard and innovation advancement benchmark), is expected to be funded by revenue from worker skill validation, and other revenue streams will be the exception rather than the norm.

As CHAMP is a novel NLP model, support from larger organizations for its incorporation in their NLP modules (such as Google's BERT) is possible, but the probability of such incorporation is hard to project and must therefore be treated as low to none.

When it comes to revenue from IP registration, some elements in CHAMP are patentable and probably worthy of patenting, but patent royalty earnings reliance is not a stable business model. Therefore, potential patent royalties will be treated as low to none.   

I apply to TPrize for the following- funding, exposure, cooperation, validation and traction.

Funding from TPrize will provide extra runway for the project on top of that I provide from my pocket, and more importantly, attract other funders, from private capital ones to donators.  

The exposure generated in TPrize will ease convincing potential regional distributors and ease user acquisition, both items of interest to me. 

Cooperation, with experts and institution, will be critical for CHAMP to succeed.

Specifically, academic cooperation will be crucial for me. It will enable me to provide validation, not only knowledge, and will increase the amount of users dramatically. 

Cooperation with commercial bodies dealing with NLP modules, such as Google, who might sponsor the project, are also of great interest.

Validation is a key element for a project like CHAMP, especially in the beginning. Skills are valuable, but employers hire by validation- if someone cannot show approval they know something, they are regarded as if they don’t.

It will take time until employers will trust CHAMP as assessment tool, and until then (and hopefully after), another source of validation can boost user base.

Good partnering academic institution can provide such validation, and hopefully one will be found following the competition.

I hope the competition, as it is high-profile, will create traction around CHAMP and make people follow it, and more importantly, talk about it.  

In addition, help with IP registration can help me build a solid patent portfolio and will be welcomed.

First and foremost, I would like to partner with universities providing both help in the research elements of CHAMP and in validation of its output, as well as validation of its learners in this way or another.

Globally, top-tier universities will be fantastic, as both the research capability and validation value of such institutions are massive. Even one such university cooperation will propel the solution quite much further.

Regionally, well-known universities in LATAM and the Caribbean are a great partner for us, as validation from them is expected to have very high regional impact on solution adoption rates. 

In addition, I look for co-founders of proper caliber, especially from academia. If any institution, or researcher in an institution, is interested in joining me, I am very willing to hear more. 

Ministries of education, especially ones in LATAM and the Caribbean, will be ideal partners in user acquisition. It is enough that a sticker with a QR code (containing a link to the solution) will be placed in each school to enable us access to millions of users. Their benefit is higher skilled population without added education costs.     

Organizations dealing with natural language processing, and mainly Google, are of great interest to us both as cooperators and as sponsors.

I think any interest in the project from them will come only in later stages, but if I'm wrong, I'd love to cooperate with them.