An Interview with Runik Mehrotra, President, AI Labs

Comment

An Interview with Runik Mehrotra, President, AI Labs

  1. Runik, can you tell me a little bit about your personal background and how you got involved with AI Labs? For sure. I started out as a technical guy, working for a large mobile application development firm. While working there, I realized that the market is super inflated and that I could make a lot more money by creating custom applications for clients about 80% cheaper than they were currently being priced. So my business partner and I started NYX Development, a mobile app development firm that worked to undercut the entirety of the mobile application market. After a successful exit, the two of us moved on to founding AI Labs. I was able to utilize a lot of my Mathematics background with AI and we both knew it was about to be a huge market. My current involvement with AI Labs is as the President. I oversee the technical team and development and work with the management team to manage the operations.
  2. Which specific markets are you targeting and why? We are targeting the institutional wealth management market, as this market is currently operating on legacy software solutions with most portfolio management being done by manual quantitative and qualitative analysis. Millions of wealth managers and financial advisors spend hundreds of hours every month creating custom investment solutions for their clients. In reality there is only so much financial data they can crunch, so many news articles they can read and so many portfolios they can create custom to their client. And actually after doing extensive market research, we found out they often slack off and sell generic mutual funds to their clients, giving them a solution that may not be perfect for them. Our goal is to be able to replace research divisions of advisory firms and automate workflows to cut overhead for these firms. FSAI essentially replaces the role of a research analyst and assists an advisory broker by creating a customized portfolio for the client, removing hundreds of hours of research. FSAI is able to analyze every earnings report for the past 10 years, hundreds of news articles, and crunch tens of thousands of numbers to build a portfolio, drastically minimizing the ability to make a mistake.
  3. As you know, there are hundreds, if not thousands of machine learning startups in existence today. I’ve spoken with several investors who are investing in as much as 5-10 machine learning companies each month right now.  What make AI Labs unique, and how do you plan to build a viable business in multiple markets with so much market confusion and competition? There are hundreds of FinTech startups as well.  And although AI Labs is not competing with most of them, there is definitely a lot of market confusion. I have seen many investors say they will only invest in a couple AI startups and some who say they will invest in as many as they can. AI Labs is different because of our approach and the market we are targeting. FSAI is viable because there is no upcoming competition in its target market. FSAI competes with the legacy solutions currently on the market, the existing software applications that brokers and advisors utilize for stock selection and investment management. Our service is driven by artificial intelligence and will be the new benchmark in financial technology. By developing a scalable automated platform, FSAI will disrupt the money management industry by minimizing reliance on human capital thereby allowing wealth managers and financial advisors to reduce operating costs and focus on growing assets and forcing higher returns.
  4. You say that you have a unique proposition in that your platform is a combination of many different aspects of AI. Can you elaborate? There’s a slight misconception right now about AI. Artificial Intelligence isn’t really a technology, it’s more of an umbrella term for powerful machine learning and computational intelligence technologies and algorithms. For some AI startups, a linear regression or a recursive neural network is enough. However, to truly emulate a research analyst, we use almost 20 or 30 of these technologies. In addition to what’s mentioned above, we use almost everything from Statistical Regression and Genetic algorithms (Particle Swarm Optimization) to Means End Analysis, and Natural Language Processing to create an intelligent system that can accurately analyze the market.
  5. Does your management team include professionals who have specific domain experience in the industry’s that you are targeting? Our team has a significant amount of past experience in both Machine Learning/Artificial Intelligence and Financial Services. Every member of our technical core team has been previously a part of FinTech startup or brokerage firm and a renowned individual in data science and AI. They have both the financial and AI knowledge to create a powerful product. Our Management team also consists of professionals in the industry. Our team includes past employees at JP Morgan, Goldman Sachs, and Fidelity, experienced RFA’s,  past executives at publicly traded companies, and previous employees in the advisory brokerage space. The concentrated team is passionate about the problem and has the technical skill to solve it.
  6. Can you explain your platform in terms of architecture and core technologies used? We have positioned ourselves as a stealth AI startup so a lot of what we do is proprietary. Especially in AI and Fintech, powerful and accurate algorithms are intensely sought after. So there’s not much I can explain in terms of architecture. What I can say in terms of our core technology is that we use a powerful machine learning platform with many elements to accurately predict risk and maximize return.
  7. Would you say you are predictive analysis platform, a learning platform or an AI platform? Please elaborate. I think we are a little bit of all three. Finance and risk prediction is a pretty difficult field to quantify and be accurate in. Our core prediction model is machine learning based but can definitely be classified as predictive analysis as well. And like I mentioned earlier, all of this is AI, we are using AI for almost every single calculation FSAI makes. This combination of all three is what makes FSAI so powerful.

Comment

Neural Networks: Applied AI Series I

Comment

Neural Networks: Applied AI Series I

 

I. Natural Neural Networks

The essential contrast between a characteristic neural system and a Distributed Processing simple of a Neural Network, is the endeavor to catch the capacity of a genuine neuron and genuine normal courses of action of neurons in the model. From the Hebbian model utilized as a part of ahead of schedule perceptions to present day neural models that endeavor to catch the biochemical strings that execute diverse types of memory inside of the same cell, the thought has from the beginning been to locate a sensible model for the neuron, and gain from usage of that model information of how normal neural frameworks may function. It has been a long hard street, keeping in mind neural systems have picked up and lost unmistakable quality in A.I., Neuro­researchers have been compelled to do a reversal to the Neural Network model, on numerous occasions, as the most moral way to deal with finding out about characteristic networks of neurons.

Dissimilar to different types of Neuroscience, Neural models don't slaughter creatures, with a specific end goal to get their neurons, they don't torment creatures keeping in mind the end goal to perceive how they will respond, they don't even include genuine creatures, rather they torment recyclable electrons by making them move through PC circuits. Something that can't even be seen, and since electrons are not thought to be alive, there is no moral purpose behind worry, aside from maybe for the utilization of power, however demonstrating is not an extreme utilization of electrical amperage, since PCs are so productive. In short there is almost no morally amiss with tormenting electrons.

The issue has then gotten to be: What are the best models for:

Neurons
Networks
Learning Algorithms

II. Artificial Neural Networks

Lately, neural computing has risen as a viable innovation, with effective applications in numerous fields as differing as finance, medication designing, geography, material science, physics, and biology. The energy comes from the way that these systems are endeavors to display the abilities of the human cerebrum. From a factual point of view, neural systems are fascinating on account of their potential use in forecast and order issues.

Manufactured neural systems (ANN) are non­linear information driven self-versatile methodologies rather than the conventional model-based strategies. They are intense instruments for demonstrating, particularly when the basic information relationship is obscure. ANN's can distinguish and learn connected examples between info information sets and relating target values. Subsequent to preparing, ANN's can used to anticipate the result of new autonomous info information ANN's mirror the learning procedure of the human mind and can handle issues including nonlinear and complex information regardless of the possibility that the dam are loose and boisterous. In this way, they are preferably suited for the demonstrating of farming.. which are known not complex and regularly non­linear. An imperative element of these systems is their versatile nature, where learning by illustration replaces programming in taking care of issues. This component makes such computational models exceptionally engaging in application areas where one has little or deficient comprehension of the issue to be comprehended yet where preparing information is promptly accessible.

These systems are neural as in they may have been motivated by neuroscience however not as a matter of course on the grounds that they are reliable models of organic neural or subjective wonders. Truth be told, a larger part of the system are all the more firmly identified with customary numerical and/or factual models, for example, nonparametric example classifiers, bunching calculations, nonlinear channels, and measurable relapse models than they are to neurobiology models. Neural systems (NN) have been utilized for a wide assortment of uses where measurable routines are generally utilized. They have been utilized as a part of characterization issues, for example, distinguishing submerged sonar carrel, and perceiving discourse. In time arrangement applications, NNs have been utilized as a part of anticipating securities exchange execution. As analysts or clients of measurements, these issues are typically fathomed through traditional factual techniques, for example, discriminant investigation, logistic relapse, Bayes examination, numerous relapse, and ARIMA time­arrangement models. It is, in this manner, time to perceive neural systems as an effective instrument for information investigation. 

Comment

Insight Artificial Intelligence

Comment

Insight Artificial Intelligence

As soon as it had been set up over 50 years ago, the AI industry has been specifically geared towards your development regarding "thinking machines"—that is actually, desktops along with human-like standard brains. The whole package, with comprehensive bells and whistles like personal, will attentively, creative imagination, or anything else.

 

But this target demonstrated quite hard to reach; and so, over time, AI scientists attended to concentrate largely on producing "narrow AI" techniques: software program presenting brains about particular responsibilities throughout somewhat filter fields.

 

That "narrow AI" operate features has sparked research that has been fascinating and also productive. It has developed, as an example, chess-playing software programs that could eliminate any man; and also software programs that could detect conditions better than doctors. It has developed software programs that will turn talk to be able to textual content, analyze genomics information, travel programmed motor vehicles, and also predict stock options prices. The particular number moves on and also on. In reality, core software program like Google and also Mathematica make use of AI algorithms (in your impression that will the root algorithms resemble people tutored throughout college or university lessons on AI).

There's cynical expressing that will after a number of target may be attained by way of computer system system, it really is not AI.. Though the deeper reality why these narrow-AI achievements amaze people is actually the way various all this progress inside design regarding particular AI tools really is from what is actually desired to manufacture a imagining unit. Each one of these narrow-AI achievements, useful while they usually are, never have however carried people quite far towards with regards to setting up a genuine imagining unit.

 

Many scientists feel that filter AI at some point will probably steer people to be able to standard AI. That as an example might be what Google's creator Sergey Brin indicates when he calls Google a revolutionary 'AI business'. His concept looks like Google's narrow-AI work towards textual content research and also linked difficulties will probably gradually bring about better and also better products that could at some point achieve genuine human-level understanding and also cognition.

 

However, various other researchers believe that narrow AI and general AI are fundamentally different pursuits. Using this view, in the event that standard brains is the goal, it will be important intended for AI R&D to be able to refocus by itself towards an original goals in the field—transitioning away from the current concentrate on extremely particular filter AI problem handling techniques, and transition toward dealing with the more challenging difficulties regarding human degree brains and also eventually brains past human degree. With this thought, AI scientists began with the period Unnatural General Thinking ability as well as AGI, to differentiate work towards standard imagining products from operate geared towards developing software program handling a variety of 'narrow AI' issues.

 

Many of the operate accomplished so far on narrow-AI can participate in an essential part generally AI research—but inside AGI view, in order to be so useful, this operate will have to be considered coming from a various view. It is actually how the crux regarding brains generally is because of your emergent structures and also mechanics that will happen in a complicated goal-achieving method, allowing this technique to be able to design and also predict a unique total synchronized conduct designs. These structures/dynamics include items all of us sloppily describe along with terms like "self", "will" and also "attention. "

 

With this in mind, considering a new intellect as being a toolkit regarding particular methods—like the ones developed by narrow-AI researchers—is misleading. The intellect need to contain an amount of particular functions that will synergize collectively so as to bring about the right high-level emergent structures and also mechanics. The person aspects of a good AGI method may well occasionally resemble algorithms put together by narrow-AI scientists, but centering on the individual and also singled out functionality of numerous method ingredients is just not awfully profitable in the AGI situation. The main position is actually how a ingredients come together.

 

Concerning specialization and also generality, inside neural is actually subtler as compared to is frequently recognized. Mental performance definitely features a number of particular tools, for instance it deals with face recognition algorithms. But these are generally definitely not your substance regarding human brains. Many of the brain's weaker tools, for instance, quite poor algorithms intended for reasoning underneath anxiety, have been far more crucial to be able to understand standard brains, while they get subtler plus more thoroughgoing synergies along with additional tools that will assist bring about significant emergent structures/dynamics.

 

Today, the term "general" inside expression "general intelligence" really should not be over interpreted. Genuinely and also entirely standard intelligence—the capacity to remedy most conceptual issues, regardless of the way complex—is difficult throughout real life. a couple of Mathematicians get demonstrated which it could hypothetically be performed through theoretical, much strong desktops. Though the tactics useful through these types of much strong hypothetical products do not have much to do with authentic products as well as authentic brains.

 

But even though entirely standard brains is just not pragmatically possible, still, it really is crystal clear that will mankind show a sort of standard brains that will should go past cures notice throughout chess software programs, information analysis software programs, as well as speech-to-text software program. You can easily get into completely new situations, number all of them available, and also produce completely new designs regarding conduct based on what we now have figured out. The man can take care of situations of the radically various nature as compared to anything at all recent before the birth—but a new filter AI system normally starts off acting stupidly as well as screwing up entirely when confronted with situations diverse from people created through it is programmer. We all mankind, took over even as usually usually are through the simian ancestral roots, nonetheless possess a knee high on Deep Blue, Mathematica as well as Google inside fluidity and also generality division. We all recognize, with a degree, exactly who and also cures usually are, and also the way we usually are related to the environment—and this understanding allows us to take care of new contexts artistically, adaptively and also inventively. And this also, My partner and i posit, happens out of your emergent structures and also mechanics that will happen inside complicated techniques that are the brains, a result of the friendships of numerous particular ingredients just a platform that will progressed to back up specifically these kinds of beginning.

Comment

Neural Networks Rethought

1 Comment

Neural Networks Rethought

                Although the word “Neural” would spark quite a biological discussion, it would also spark a conversation about Artificial Intelligence and Neural Networks. Since the dawn of intelligent beings, the humankind has wondered about how our brains work. And although we have gotten quite far in explaining the reactions that occur in the brain, we are nowhere near being able to fully understand the brain. Similar to the brain, we have barely scratched the surface of neural networks and their possibilities. An artificial neural network (or simply neural network) is a encoded network of nodes which act as "neurons" while mapping out a network of nodes similar to the human brain. The simplest definition of a neural network is stated by Robert Heicht Nelson, the inventor of the first neurocomputer.

               ". . . . a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs."

 

           Neural networks consist of at least 3 node-layers. The input layer is responsible for identifying and defining the input data. The data then reaches "hidden layers." Such layers have the name hidden layer as it is quite difficult to distinguish each hidden layer. The hidden layer transfers the input data to viable output data. There can be infinitely many hidden layers. The final layer, known as the output 1 layer, is responsible for outputting the new data. This system of networks have many applications in developing algorithms or recognition software.

1 Comment

Comment

What If Apps Could Make Themselves?

AI Labs software program is a new form of mobile app development where no programming on the consumers part is required. The process starts with a consumer logging into our software system and deciding what they want in the app by selecting purpose of the app and needed features from a given list such as calendar, store etc. Then they choose a design scheme such as modern and sleek or match my photos (matches color scheme to the photos uploaded), and they upload a few photos/logos. Our software will then add all of your specifications together, and show a series of designs to you. You will choose which app design you like the most, you then select it and type text, and add content in the needed areas. Then when complete the app design will automatically turn into a .app file to be published on the App Store. If the consumer changes their mind about a feature they have on the app or wants to add a new feature they simply log back into the software and add the feature by clicking on it, then upload their (photos, videos, customized text, etc) and click update the then app is instantly updated. Therefor the app is customizable to each user, as well as unique to the selected design of the app. Via this process you do not need to drag and drop, use templates, and hire programmers. Just imagine making an app in minutes. 

Comment

Comment

A Preview Into The Future Of AI

    Though it has taken 150 million years to reach current day, the intellectual journey was not gradual in a linear sense. If one was to plot significant events occurring throughout human existence, Mankind’s ability to construct new ideas follows a logarithmic path, and is rapidly approaching an asymptote, or technological singularity. This singularity event has scientists both supporting and rejecting the concept of an imaginative plateau; the largest topic discussed is Artificial Intelligence (A.I.). When this technological singularity is reached, it is hypothesized that man’s greatest creation, an artificial sapient being, will supersede human brain capacity. Because of this drive to exceed the technological barriers, we have never been closer to developing a functional A.I. However, more companies are creating AI as a vast list of commands, known as “frozen” intelligence, rather than developing true intelligence. This “frozen” intelligence never has the capacity to become an A.I, as we can never program an infinite list of commands. However, true intelligence, an intelligence able to program itself and learn, can become an A.I.  An AI able to emulate human thought, human experience, and human emotions.  

Italics -Previous Passage 

Comment