Monday, February 29, 2016

Interview with Philip Thomas, CEO/Founder of Staffjoy

What does your company do?
Staffjoy automatically schedules people so that they work when they want and the business saves money. What we are trying to do is create flexibility for workers and businesses to help them thrive. The first place where we are starting to do that is the scheduling of shifts for hourly employees.

What caused you to start Staffjoy?
Staffjoy started as my senior research project at Washington University in St. Louis. I had been doing a lot of applied mathematics work, and for my senior project I had decided that I wanted to try to make something interesting and novel, rather than partner with a corporation like a lot of other students did. I worked on developing new techniques for scheduling workers that aimed to put them into the shifts that they prefered. My senior project was nowhere near where Staffjoy is today, but my interests began with that project, and I continued working on it on the side for a few years before my friend Andrew Hess joined in to work on turning some of these ideas into real production systems. And then shortly after that we closed our first customer. In October of 2015, Staffjoy transitioned from being a part time pursuit to being our full-time job. So we are now a team of three people with number four starting in a week and a half.

What makes Staffjoy a AI start up?
So I originally struggled with the question of whether to classify Staffjoy as “Artificial Intelligence” or “Operations Research”. In general, I have moved towards calling it more of an “applied mathematics” company. However,, external sources have often referred to us as an AI company. For instance, LAUNCH Festival named us a “Top 10 Emerging Startup in Artificial Intelligence and Machine Learning”.. And this is something that I think is an interesting question - What defines Artificial Intelligence? I wrote a blog post on the topic, but the question kind of becomes "Is Artificial Intelligence defined by the work the agent does, or by how the agent does the work?". So what we're doing at Staffjoy is automating the decisions that humans make, and we are able to replicate or outperform humans in terms of the quality of those decisions. But what makes Staffjoy more in the realm of Operations Research is that there is a measurable quality to some of these answers. For instance, there is a minimum number of labor hours needed to meet business needs. In that case it is definitely more of an Operations Research approach, because you're doing convergent optimization. I think that this tends to be a little bit outside of the scope of academic artificial intelligence.

I tend to think of hard Artificial Intelligence as a neural networks and science that attempts to replicate human thought patterns in order to do work. But Staffjoy is mainly an Artificial Intelligence company in the sense that it replicates decisions made by humans and does so better than a human. But the question still remains to be seen, "Do we define Artificial Intelligence as the work done, where it literally makes decisions that are typically done by an intelligent human? Or, do we define it by how it performs the work?" We don't really outwardly talk about the work we do at Staffjoy as “Artificial Intelligence,” but others have described what we do as just that.

If I understand correctly, the goal of your AI or algorithm is to optimize staffing?
The first problem we worked on solving was generating shifts of dynamic length for a group of workers that meet business needs. A business has varying needs throughout the day. If we are looking at something like a coffee shop, they know that traffic levels will be low at 8 am, slightly higher at 9 am, and at a peak at 10 am, and then start to decrease from 11 am. So there is this need to vary staffing levels because if they maintain the same number of workers throughout the morning then that means that either at the beginning of the morning when things are slow they have more workers than they need so they are over-scheduled and paying more money than they need to. But, it might also mean that at 10 am when they are at the peak traffic, the number of workers might not be high enough,  so they may be understaffed and unable to provide service at the level of quality that they need.
That's part one, looking at the demand of the business, which can be forecasted by looking at historical sales numbers, and matching scheduling to dynamically increase and decrease throughout the day based on the needs of the business. So take forecasted demands, convert it to minimum staffing levels, and then we optimize schedules to meet as precisely as possible to meet  varying demand levels throughout the day.
This is subject to a lot of rules. Let's just talk about some of California's compliance rules. Depending on how a “workweek” is defined, in general working 6 or more days consecutively in a workweek means that somebody qualifies for overtime pay. If they work more than 8 hours in a workday then they are subject to overtime pay. If they work more than 40 hours in a workweek, they can be subject to overtime pay. So, what starts to come down is this pretty basic model of match labor to business needs, but then you layer on top all of this compliance and it becomes significantly more difficult as you have a basic, quantifiable objective function -  “minimize my labor costs” -  but then you have to make that subject to a many different compliance constraints. And then there's real world constraints - maybe one of the workers is a college student and is only available to work on Monday, Wednesday, and Friday or Saturday after 1 pm. As you start to layer in all these different constraints it starts to become extremely difficult, not only to match the business needs, but also minimize the labor costs. At the core that's the mathematical problem that we are solving for many workers is scheduling them to work when they are available to, subject to all of these constraints.
What we're working on doing right now is releasing an even more advanced version of this algorithm. For a given business there may be billions of possible schedules that are feasible and meet the constraints. Out of those billions of schedules, we generally estimate that a few thousand are optima. What we try to do is, after we find the threshold of optimality, we try to search for basically what amounts to canonical optimal solutions that still meet the business needs but try to put people into the shifts that they prefer. So what do I mean by prefer? We actually try to quantify that right now by asking workers for their preferences. Those preferences may be "I like to work mornings,” "I like to work afternoons,” things like that. And we've developed an economic model for scoring shifts for these workers.
This is all starting to sound really complex, so taking a step back, what are the actual benefits to the business and the workers? The benefits to the business are that they have optimal staffing levels and on average our data is showing that we can decrease overall labor costs to these businesses by about 10% through more optimal staffing. But then on the workers, what do they gain? Well they gain flexibility, because they are able to have more control over when they work. So, a morning person will not get scheduled in the closing shift, making them happier and more productive. We're able to start to collect a lot of data that can make even more knowledgeable solutions in the future.

Can you describe how you go about building a better algorithm?
When we originally wrote the first version of the Staffjoy algorithm, it was a basic branch and bound algorithm implementation and we were so excited because we we had all of the constraints programmed in, we had the objective function of minimizing labor costs and it worked. And then we got data from an early user. They sent us the actual inputs for 30 of their workers, then Andrew and I spent hours turning the spreadsheets into code, because we didn't have any automated way of doing that yet, and we hit “run “on this 32-core EC2 Machine, an Amazon server, a really big one, and we stepped back. We waited, and we waited. After three days and about seventy dollars in pure server compute costs we finally got an answer and it was correct. But that was extremely expensive to compute because the cost of the server was more than the money we would have hypothetically been saving some of these businesses.

So then we went back to the drawing board. And a lot of the work that we've done recently has focused on the realization that viewing the scheduling problem as a perfect convergent optimization solution isn't always the best for both parties because, by looking to be within a few percentage points of perfect convergent optimization, we're able to significantly speed up the calculation using a variety of different techniques that we've developed. Our work has focused on sub-problem generation,heuristics ,and tuning production models. The basic core thing we've realized is that an NP hard problem, when you split it into two separate problems, the sum of the time it takes to solve the two child problems is significantly less than that it would take to solve the parent problem. That's a lot easier said than done, though.

Why do you think there is so much excitement about AI and start-ups?
Looking at our customers, very specifically, why is it exciting? It's exciting to them because our algorithms decrease their labor costs by about ten percent. Meaning, we have one customer that has been using us for a long time, we compared their manually created schedules with our Staffjoy created schedules and we decreased this company's labor costs by eleven percent without decreasing the quality of service. In fact we actually increased the quality of service because it turned out that for about four percent of the time throughout a week they were under-scheduled and losing business.
Labor is about a quarter of the costs of most businesses, so a ten percent decrease in labor costs results in a massive increase in overall business profitability because labor is such a high cost. So that's Staffjoy, we are able to do really interesting things with specialized algorithms that replace human decision makers, and along the way save money.

But in the broader perspective, why are these businesses so eager to adopt AI? I think we are seeing a movement towards businesses feeling comfortable pushing decisions to computers. For instance, one of the things we've had to do is un-train a lot of managers from feeling the need to push an “accept” button or a “publish” button on some of the outputs of our software. That's because they've started to shift more confidence into the ability of computers and specialized algorithms to make decisions that don't need necessarily as much oversight. I think this is a special time because this means that it opens up the ability for computers to be able to replace humans and be able to run by themselves with a degree of autonomy. I think it's just fascinating that technology is moving from being a tool to gaining its own credibility across fields to make decisions autonomously. The cool part of Artificial Intelligence is that we can talk about what is the difference between soft AI and hard AI, like a neural network, but the fact that people are finally embracing technology to make decisions for them in environments like businesses where the costs and ways of doing things are often ingrained in the mindset of the people running them is really fascinating. And I'm starting to see in my own life the hypothetical Turing tests everyday as I email people and then I realize after the fact that some of the assistants were human, others were robots, and I had no ability to distinguish between AI assistants and human assistants myself. I now see that people are becoming more accepting of technology not just as a tool or an app, but asba decision maker

Monday, February 15, 2016

Interview with Borui Wang, CEO/Founder of Polarr

Borui, who are you?

I'm the founder and CEO of Polarr. I graduated from Stanford studying Human Computer Interaction and Artificial Intelligence. I love photography and subjects that link art and technology together. 

What does Polarr do?

Polarr makes thing pretty at scale. We build the best tools and services to beautify the world's images. 

What caused you to start Polarr?

I want to work on something I care about and I love building tools for others to enjoy and use. My passion for photography makes it difficult for me to get bored working on photography related concepts. I think Polarr's mission makes me feel more connected to my passion and the people who use our products and services. 

What makes Polarr an AI startup?

AI is everywhere. Is there any new software startup that doesn't have an AI component in their road map? For the real though, our team has several experts in ML and AI is a huge component in computer vision and computer image analysis. To make things pretty at scale, we've developed a range of algorithms to understand how people edit image and try to incorporate machines to better assist people and eventually replace people for most mundane tasks. None of these modules are in our products yet, but they will slowly show up in our future product offerings soon. 

Describe Polarr's AI.

We are in the process of building systems to enhance images automatically at massive scale, with high quality data learned from millions of pro photographers using our photo editor. It will take a lot of tuning and proper data ingestion and pruning, and we believe we've nailed most of the process and workflow now. 
How do you measure and communicate the quality of your AI?
We compare our machine results with human results and mark visual quality from a group of experts, and we have an internal system built just for visual benchmarking. 

Can you share something awesome that your AI has done or that you have been able to do with your AI that would surprise most people?

How about I tell you we know how likely your photo is going to be liked by a lot of people or even win a photo competition ? : P
When you talk to investors, do they care about your AI or do they care that you are solving a business problem? How about customers? How about the media?

The investors obviously care a lot about the AI component as that's our core tech and IP of the company. However, they won't care about the AI part so much if it doesn't eventually solve a massive business problem.

Why is there so much excitement about AI startups?

AI is eating the world. It decides what we eat, who we date, and where we go for vacation. I think the opportunity is vast and profound and people haven't completely grasped the potential and what it is yet, therefore a lot of excitement for the unknown. On the other side of this - the humanity and social science research of AI is also a relatively new subject and is barely explored. In movies and pop culture we see the growing interest in robotics and machine intelligence - it seems to be something humanity in general cares and wonders about. 

Where do you go to get information about AI startups?

I read hacker news as much as I can to stay on top of discussion in startups in general and often time AI startups in general. I also pay attention to papers from top conferences in AI, especially in NLP, Deep Learning, Computer Vision and pay attention to what researchers spent their time on. A lot of AI startups grow out of research. 
Make a prediction about the future of AI startups.

There will be huge disappointments because the public expectation of AI is often different from what it is and what it actually could do, but there will eventually be a major leap and break through that surprises and even scares people. For startups, funding will still be sufficient as long as the AI is solving a real world problem.

Monday, February 8, 2016

Interview with Max Versace CEO/Co-founder of Neurala

Who are you?
My name is Max Versace. I am the co-founder and CEO of Neurala, which was started in 2006. Myself, Anatoly, and Heather, while we were studying for our theses in cognitive and neurosystems. We are launching a company based on artificial intelligence to change the way robotics is done today.

What does your company do?
Our company designs artificial brains for machines. It's a pretty broad task, so we specialize in designing artificial nervous systems, or emulation of the nervous system for ground-level drones, automotive or whatever has a machine in need to operate in the real world. Basically to be able to perceive this world and interact with it and navigate in it.. And that's basically the goal of getting a machine ready to perceive and move around, imitating aspects of how animals achieve these capabilities.

Can you say more about your AI?
Our AI starts from a scientific background. We started out for three years studying the brain, brain areas and brain competencies ranging from visual perception, memory, auditory perception, spatial navigation, and our journey started from neurobiology and mathematical modeling of these capabilities in software, until at a certain point we realized that we were actually solving problems for a huge industry, which was robotics, but now the term 'robotics' has expanded to encompass drones and automotive. So in essence, we're solving a problem for a huge variety of machines that up until today have been driven by humans, but today we want them to be driven autonomously and serve humans, and that's our unique approach. We design AI by basically mimicking the way it has been solved by biological systems millions of years ago.

How did you obtain the expertise to build this type of AI?
We obtain it by myself, having two PhDs in the topic, and the other co-founders one PhD each, and there are more expertise in the company beyond the co-founders. But I've been studying the brain and how to emulate it in math for the past 20 years. And there are different ways in which you can build the stuff up with AI. You can download the software package from the internet and try to make it work and have small problems, or you can build these algorithms over the course of decades and really understand them from the ground up, and have under your belt several publications, tens of publications with your colleagues on how to produce these algorithms really from scratch. And I belong to the second species.

How do you measure the quality or the performance of your AI?
There are ways in which you can do this, and there are many ways in which it is not possible to do it. And when I say this, let me qualify the statement. Sometimes you can prototype your AI against data sets, and in many domains from vision to audition to classification and so forth, there are many data sets that you can test your AI against before even deploying it into the real world. We do that of course, and pretty much everybody either does it or should do it, but at a certain point there is a leap of faith when you're leaving your data set validation and you're going to the real world. And no data set will capture the real world. So really there is no way to validate your AI other than deploying it. And that's where the pain starts, because the cost of making sure your AI works grows really much quicker when you step out of that very well-defined and confined benchmark where you can test the software.
On a day-to-day basis, what are you doing to improve your AI?
That's all we do. We do AI on a day-to-day basis so we improve from the day before, in a sense. We have several things going on. Neurala is a company that is an exclusively AI company, exclusively software company. So we improve the software, whether the software is software we already have shipped to consumers in the form of apps, and we have a couple of these already in the market, whether it's software that we are designing for nav in the Air Force which is still R&D software, or whether it's software that we are integrating to our customers, in this case this can be a robotic company, a drone company, or a semiconductor company.

Can you share something awesome that your AI has done that would surprise you or has surprised other people?
Yes, the most surprising things we have done in the past few months. We did a demonstration at NASA, where for the end of our project phase with them we had to demonstrate a robot that is able to navigate its environment while maintaining the sense of its position in space, classifying objects, placing them in a map. It was pretty much a 'wow', not only for NASA, but even for us when we saw the whole thing all at the same time operating in the real world.
So that was an achievement we did in the summer, and in the fall we took a little piece of that algorithm and we put it in apps that are today available to consumers. So, small step for humankind, or, for neurologists; a big one for humankind, if I can adapt Armstrong. We are bringing the power of AI into the hands of consumers. We are doing this step by step, but I think the 'wow' that we had at the end of 2015 was the first step of productizing this technology.

When you talk to investors, do they care about the AI, or do they care that you're solving a customer problem?
There are certain people who care about AI and they're excited about it. Others, they don't really care. So I find all sorts of people, and it's probably no surprise to you that I tend to feel that people are more excited about the technology.

Why do you think that there's so much excitement about AI now?
I have elaborated on this several times in the past few years. I believe that robotics is a very tough challenge and smart machines are a very tough challenge, but today the things that are needed for machines to be smarter are all here, and that's why we are seeing this excitement. The first one is cheap robotic bodies, bodies that are able to be produced very, very cheaply, and today robots cost tens of dollars, and you can buy them at BestBuy. The second thing is processing power. It has come down and GPUs, or graphic processing units, that once upon a time costed thousand of dollars, and were only available in workstations, today are in people's phone. And the third ingredient of course is AI.
So all these three ingredients need to be present simultaneously for a revolution to occur and that's what we are seeing today. And we started to first see this trend in 2006 when we started Neurala, and we tapped into the idea of running big networks or neural networks from GPUs.

Where do you go, or do you suggest people go, to find out more about AI and AI startups?
Ihave a few places where I write some of my stuff. I have a blog called Neurdon  which is an intersection between 'nerd' and 'neuron'.I have a lab at Boston University called Neuromorphics Lab where we have tons of information articles. The other thing is to read scientific article about artificial intelligence, read new articles about AI, this might be little bit too technical, but I think going down to the source of a scientific papers is important.

Monday, February 1, 2016

Interview with Ankur Modi, CEO/Founder of StatusToday

Who are you?
I'm the CEO and Co-Founder of StatusToday. We are a data driven, cyber security startup that is looking to solve the challenge of insider threat in enterprises and businesses.
Can you say more about what StatusToday does for its clients?
At StatusToday, we’re trying to address the current cybersecurity crisis. The traditional approach to security via exclusive perimeter protection on systems via passwords, firewalls, access control and even encryption doesn't work.

What works is the ability to use contextual knowledge from a variety of sources to better understand activity. We use three key sources of knowledge for our analysis. First is psychology, or more specifically organizational behavior. Second is the data from large scale analysis of activity logs, i.e. data that is already collected in most organizations and IT systems. And third is a human centric activity model we have created to better understand how users interact in distributed IT systems. The combination of these three, has allowed StatusToday to better understand systems and natural behavior, to predict deviation and detect anomalies that are otherwise too human to detect.
Our charter is very simple. Instead of locking down systems beyond the point of usability, we're trying to advocate an open system model, where malicious activities are detected and flagged even if the user is compromised.
On our way, we’ve researched most of the major data breaches in the last 5 years; from NSA, Sony to Ashley Madison. What we've found is that, while the breach might originate outside the company, the eventual medium of attack is almost exclusively via internal employees. Humans are, by far now, the weakest link in any security system. We help companies detect such incidents, explore possible impact and then provide the tools necessary to understand the damage. If you look at recent breaches in the industry, most companies were unable to answer the most basic of all questions: "What information was actually compromised?". At StatusToday, our solution is able to help them answer this fundamental question promptly. Detect if something happened and if so, identify the scope of the damage immediately.
What caused you to start StatusToday? What's the origin story?
I worked within the emerging field of data science at Microsoft for the last 5 years in Ireland and Denmark. Most recently I was behind the Microsoft Office Store platform managing the ocean of data that was generated from user behavior in order to find out how to enhance popular features.
I reached a defining moment at Microsoft when I understood that, the power of big data doesn't lie in the data itself, it lies in the way you analyze it. Deep insights come from an external understanding of the data – sometimes via an in-depth understanding of the user, sometimes via human psychology, sometimes via user behavior.

I decided to leave my position at Microsoft to start the company and met Mircea out here in London at Entrepreneur First. I started knowing that security is an industry that was long overdue for a fresh approach. The big data and human centric AI based approach I was about to take, was relatively new and teaming-up with Mircea, who used to head up security in large enterprises, was an ideal combination.

Insider threat by itself was not our initial focus. We started experimenting with large industrial datasets, the most noteworthy being one from DARPA, the US Department of Defense on malicious insiders. We soon found that there was a lot of power in the human-centric methodology we adopted. That's when we started StatusToday.
Tell me more about the AI that you've developed.
The AI that we've developed is quite unique, and I know that this is something every tech startup would love to say. One of the reasons our approach is truly unique is that we are able to connect distributed sources of information and handle them in a source agnostic way. The data could come from a Windows file server, a cloud-based API (like Google or Microsoft) or even service providers like Dropbox or Salesforce. At the same time, it could come from a custom backbone application that a company might have written themselves. The ability to collapse all of that into one unified source of information is huge because that gives us the power to run our AI algorithms globally. The methods we developed have the ability to renormalize all these data sources from an object-centric perspective to a more human-centric view. For example, we would to be able to see that a certain entity, has logged on to a machine, then download an unusual amount of information from Salesforce CRM, and finally sent a bunch of these via email. Then, it edited a couple of company files on Dropbox and finally left the company premises. Such a malicious chain of events, would traditionally sit in different silos of security, that we are now breaking through.

To be able to bring them together and to identify anomalies on top, can have massive implications in terms of being able to detect what is normal and what is not. That's what we’ve started building. What we have today is an advanced time series analysis that has the ability to observe subtle cues in behavior. Based on even limited sources of data, you can now infer, whether a certain individual is more introverted or extroverted.

Now fundamentally, when an introverted person gets hacked or goes rogue, there's a lot of outward activity. That change in the basic signature of the person's behavior doesn't require massive learning in terms of supervision. The change can be triggered quite accurately based on organizational psychology. That's what part of our AI does, to understand the object and the user contextually, and identify what is normal.

Our AI is able to look at several aspects of organizational normality, from time of day, type of systems, machines, activities to normal user behavior. When there's a significant change in such aspects, it is able to alert and say, "The sequence of events is quite suspicious.  The user might have been breached/compromised." That's a very strong indicator of what we call, an incident. This can then be investigated or it can then be analyzed, depending on the respective event that's in question.
How do you measure and communicate the quality of your AI?
We use a combination of approaches. We are running several pilots with large enterprises, to monitor the events that we detect. One of the simplest way of measuring an AI is to monitor actions taken as a consequence of it’s results. Measuring the AI itself in terms of its accuracy or speed is foolish in a business context, where the results matter more.

For us, the measurement of quality is not on the actual AI, but on the scenario it attempts to capture. For example, if we’re out helping a law firm, and our goal is to prevent rogue insider incidents, then we measure our AI on how many valid events get triggered that result in a tangible change within the organization. That could involve making changes to the internal security in the best case or mitigating an actual data breach via an identified incident in the worst case.

An AI should be monitored via the quality of the alert that it generates. In our case this is written in terms of risk potential. We've made it quantitative by capturing attributes like incident frequency, severity, risk potential and noise.

One of the problems with AI startups is that tuning up the accuracy even by a little bit often results in a flood of false alerts. Even for the highly sensitive enterprises out there, be it financial services or legal firms, getting a few hundred alerts a week, no matter how accurate, is not an option. Startups and solution providers need to start thinking in operational 80:20 terms. The real problem to solve is, "How can I leverage AI to tell businesses, the top five things they should take action on, that will the most impact?" Combine accuracy with possible impact to create prioritized and actionable alerts. A key member of our product team, Mihai Suteu who has an expertise in smart AI driven approaches to time analysis, has been looking at precisely this.

If we send you five alerts a week, I want to ensure that they all have a high potential for damage and you're going to do something about each of them. If we can manage that, then our AI is successful.  This is a very important measurement because it allows us to not get overly focused on the algorithm, which might be really cool or great, but rather the problem we are trying to solve.
Can you share something awesome that your AI has done that, either surprised you or would surprise most people?
"If we have a global view of everything that's happening, we are able to say when an engineer is acting like a salesperson or a salesperson is acting like an accountant." As much as people like to think otherwise, the truth is that we are actually very predictable. We all act within certain well observed parameters.

A typical engineer, as unique he/she might be, is similar to the other engineers while being fundamentally different to say a salesperson. We expected to observe this predictability since the beginning, and eventually just recently we did.
A bi-product of our approach is that we often find non-insider threats which usually lead to system optimization. In one particular case, to give you an example, we noticed a large amount of network bandwidth going through a single user account against all normal baselines. This bandwidth usage was quite rhythmic and robotic, on specific days of the week. Upon investigated, what we effectively found was, this particular user had admin access and was using a misconfigured bandwidth monitoring service to check if their web services were up or down.

The impact was that this service downloaded the full contents of the large server every 5 minutes, throughout the day under the user’s explicit credentials. From our systems point of view, this user was downloading unusual amounts of data at regular and sustained intervals in unusually automated manner. Our measurements indicate that about 30% of all the bandwidth going through that server, was to this one misconfigured service appearing hidden in plain sight under the authorized user’s activities. There's little to no monitoring on authorized access in most systems across the world. By being able to cap this out, we were effectively able to identify a way to reduce the bandwidth on this server, by 30% overnight.

Anomalies in authorized access underpins the new age of security products. This is what insider threat is all about. Insider threat is not just about bad users, it’s about hacked and compromised users who are unknowingly being used in most large scale data breaches and leaks today.
Give me your thirty second sales pitch.
We are StatusToday and our goal is to protect businesses from insider threat. We do it using multiple patent pending AI approaches that learns human behavior within organizations, to flag unusual activity. By providing organizations with the global visibility they need, we enable them to detect, investigate and mitigate any malicious activity that can cost millions in potential damage. You can reach me on Twitter.