Podcastskeyboard_arrow_rightSDS 447: Commercial ML Opportunities Lie Everywhere

58 minutes

BusinessMachine LearningData Science

SDS 447: Commercial ML Opportunities Lie Everywhere

Podcast Guest: Michael Segala

Tuesday Feb 23, 2021

Subscribe on Website, Apple Podcasts, Spotify, Stitcher Radio or TuneIn


In today's episode with Michael, we discussed how GPU’s can be used to accelerate all mathematic operations, how we can further apply ML models throughout life, data science in the private sector, soft skills you need, the biggest government policy holding back machine learning, the biggest commercial ML opportunity of the coming years, and more!


About Michael Segala
Michael Segala, Co-Founder and CEO of SFL Scientific, has years of experience leading projects that apply data science and mathematical modeling to solve complex problems. He specializes in working with corporate transformations involving data strategy, growth, automation & cost reduction, performance improvement, and organizational effectiveness through machine learning and predictive analytics. In addition to his client leadership role, Michael leads SFL’s effort to engage industry leading partners such as NVIDIA, Microsoft, and Amazon Web Services. He advises clients in healthcare, life sciences, pharmaceuticals, financial services, consumer products, retail, telecommunications, energy, and transportation.

Overview
Michael runs SFL Scientific, a data science consulting organization that he defines as having all the fun of working on projects for other companies looking to solve challenges. The group has been the Nvidia Partner of the Year, a huge distinction for the work SFL does. Their goal is to solve issues in a novel way, which makes it perfect for utilizing the power of ML. Today they work on things from surgical transplant diagnostic issues to drones and autonomous vehicles, and beyond.

SFL is full of former STEM professionals which makes healthcare an easy fit. They aim to offer novel solutions to traditional problems. An example is total heart transplants in children, which is, in many ways, archaic in its approach. Michael describes it as more of an art than a science, which is where SFL comes in. They want to take it from art to science to take much of the guesswork and natural skill out of the process. This isn’t to say transplants are going blind, they have facets and stats they utilize to determine much of the work of transplants, but SFL offers more concrete data points to optimize the choices of organs to transplant and whom to transplant them to. This ultimately could save lives and cut down the 30% of organs out there that are wasted because of lack of sophistication in transplant matching. In the private sector, Michael sees a continual problem of lack of belief and trust in ROI when it comes to investing in ML. AutoML might win a Kaggle competition, but it can’t help investigate nuanced issues and solutions that help companies and their employees grow. Michael describes AutoML like a calculator - it’s a tool but not itself the solution to a novel challenge.

In the public sector, companies not out to make a profit, are lagging in the adoption of technology. The US public sector isn’t nimble — thinking about programs in a matter of decades and losing the progress of tech that can happen in a matter of months. SFL has tried to help to adopt private-sector technological solutions into the public sector in situations that are mission-critical. Michael is passionate about doing work “at the edge”, where access to tools is limited and tools need to be able to function autonomously. SLF sees this kind of work as incredibly important for the Department of Defense who needs drones and unmanned vehicles that can potentially function on AI. One of the issues we don’t have the human resources for this work is the lack of capital and revenue available when someone can make 10x in the private sector. But, the people I personally know who work in this sector are incredibly passionate, incredibly skilled, and have job security across decades because of the slow and steady pace of progress.

Michael describes his usual every day as “crazy”. He describes it as a bifurcated focus: how do we build an organization and how do we have fun in work execution? Michael focuses on the former more often throughout his day, where he networks and “sells” their work which is deeply technical and incredibly varied. Michael’s responsibility is keeping up on all trends in the tech space and bringing them back to the team that can then apply them in the solutions they’re seeking out for their clients. The good news is, SLF is hiring across many positions and Michael shared what they’re looking for. That’s people who can wear multiple hats and be incredibly adaptive in research and approach. You need curiosity and the ability to interact with clients as well as technical skills in Python, machine learning, and general data modalities.

We shifted into looking at Michael’s professional journey. He participated in the work at the CERN's Large Hadron Collider, working in data science in an academic setting. He was part of the analysis team that helped confirm the discovery of the Higgs boson. After his academic career, Michael wanted to take these skills and knowledge to real-world, on the ground problems, taking the same techniques that solved fundamental physics problems to solve problems in medicine, farming, and other seemingly mundane issues.

Michael is excited about the booms in information and data sharing that can make difficult problems easier to tackle. COVID-19 has pushed some of this work, showing the need for rapid information sharing and the need to reorganize and reevaluate the red tape we put around data sharing when lives are on the line and the clock is ticking. 

In this episode you will learn:
  • SFL Scientific [4:20]
  • SFL’s example work [10:55]
  • Public sector vs private sector work [20:28]
  • Michael’s day-to-day [30:18]
  • What is Michael looking for in the people he hires? [33:38]
  • Michael’s career journey [41:39]
  • What is Michael excited about for the future? [48:38] 

Items mentioned in this podcast:
 
Follow Michael:

Follow Jon:

Episode Transcript
Jon Krohn: 00:00
This is episode number 447 with Michael Segala, CEO of SFL Scientific. 

Jon Krohn: 00:12
Welcome to the SuperDataScience podcast. My name is Jon Krohn, a Chief Data Scientist and bestselling author on Deep Learning. Each week, we bring you inspiring people and ideas to help you build a successful career in data science. Thanks for being here today. Now, let's make the complex simple. 

Jon Krohn: 00:42
Welcome to the SuperDataScience podcast. I'm your host, Jon Krohn and I am honored to be joined today by Dr. Michael Segala who, when it comes to real-world data science applications, has perhaps the widest breadth and deepest depth of anyone I've met. He's also remarkably adept at reeling off several detailed, articulate examples to bolster any general point he's making in conversation, I was blown away. 

Jon Krohn: 01:09
Michael has a PhD from Brown University during which he contributed to the biggest Physics finding of the millennium thus far, the discovery of the Higgs boson. Moving into industry after his PhD, he founded an AI and data science consulting firm that has grown into an elite organization, including being recognized as Partner of the Year to Nvidia, the GPU juggernaut, now several years running. 

Jon Krohn: 01:35
During this episode, Michael fills us in on how GPUs can be used to accelerate not only deep learning models but all mathematical operations, how humankind is today only scratching the surface of the opportunity to apply machine learning models in meaningful ways that will dramatically improve both lifespan and quality of life, the key differences between applying data science in the private sector versus in the public sector, the soft skills in each of the three distinct fields of deep learning you need to have at least some expertise in to work at an elite AI consulting firm, the one big government policy that's holding back machine learning innovation and that may change because of COVID, and the biggest commercial machine learning opportunity of the coming years. 

Jon Krohn: 02:20
This episode should be of great value to anyone who's interested in enriching their understanding of how machine learning and data science can be applied to make a massive impact in the real world. 

Jon Krohn: 02:37
Michael, welcome to the show. It's such an honor to have you back on the SuperDataScience podcast. It's been a couple of years. I can't wait to hear what's been happening in those years. How's your day going so far? 

Michael Segala: 02:50
It's going excellent. It's been busy. First month of the year has been very busy and hectic but all things considering, it's going very well. How are you? 

Jon Krohn: 02:59
Very well, thank you. It's super cold in New York at the time of recording. We had the first lasting snowfall last weekend. You're in Massachusetts, right? I imagine you have lasting snowfall for months. 

Michael Segala: 03:16
Yes, until about end of March. You don't put away the shovels until mid-April and then we can have some fun once May hits. 

Jon Krohn: 03:25
Nice. Do you get into winter sports up there? 

Michael Segala: 03:29
I do. I have skied my whole life, and then this past winter, I've convinced myself I was going to learn how to snowboard, but not in a way where I was going to take lessons and do it responsibly. I was just going to buy a snowboard, get on the lift, go to the top and snowboard down, which I did and it was inexperienced to say the least. Now, I'm going to take a step back and say, "Well, I probably could use an hour lesson to learn how to snowboard." Yes, I've skied in my whole life and now, I claim to fall down on a snowboard. 

Jon Krohn: 03:59
I think that's a great idea. I have an identical story. My butt has never hurt so much as the day I tried to snowboard. 

Michael Segala: 04:06
Yes, but I've committed because I bought it so I was like, "Now, I've committed to this process for the whole winter." 

Jon Krohn: 04:12
Nice. I look forward to hearing how that journey goes. How is your career journey going? For those of the audience members who weren't listening to your episode from two years ago, tell us a little bit about what you do. 

Michael Segala: 04:28
Of course. I had the privilege of running a company called SFL Scientific. We are, geez, about six years old at this point and we are a data science consulting organization. From day one of the company through today and into our future, we've always taken a very clear business stance of forming SFL with this idea of being a consulting organization, which in our space is having all the fun of working on all these cool projects for other organizations that need outside help and expertise to solve their challenges. 

Michael Segala: 05:03
I have a team of really great data scientists and AI engineers that spend all day building algorithms in large-scale data engineering environments and platforms for our customers that range from top Fortune 5 to the military and everybody else that needs that real expertise in data science skills, and we get to supply that for them. It's a great opportunity to work on a lot of interesting and novel projects. 

Jon Krohn: 05:30
Nice. As a part of that, you have been distinguished in a number of ways. I think perhaps the most notable is that you've been the Nvidia Partner of the Year for AI services a couple of years running now, which is a huge deal for people who aren't aware of Nvidia, which may not be a majority of listeners. Nvidia is arguably the premier chip maker today. Sorry to the Intel folks listening, but they've really taken off with GPUs which were originally probably intended primarily just for rendering graphics, for example, for playing video games, for creating videos as a video editor, but they've taken off for AI and for mining Bitcoin. I imagine more on the AI side, I'd love to hear the work that you're doing with Nvidia, Michael. 

Michael Segala: 06:30
Yeah. If you think about Nvidia's perspective, the data science space really shifted call it 2014, 2015 which just happened to align when we started the company, so it was very good timing, and bringing GPUs to the market opened up the whole space of deep learning. It's not that deep learning didn't exist before, but the computational horsepower to solve some of these networks, that was really the novelty of why GPUs were interesting and stay very interesting. 

Michael Segala: 07:01
For us, you have to realize myself, my co-founders, and pretty much all of our employees, we came out of STEM-specific academic backgrounds. We were all PhDs doing lots of fun type of work but a lot of hard novel, I'm going to call it R&D work. Not R&D like you're sitting at a lab bench pipetting all day, but R&D in the sense like it's novel in the sense you want to build something with AI at its core. All of those projects mapped really well to saying, "Well, how do I solve this in a novel, interesting, hard way where deep learning can start really adding that extra layer I can get now more accurate, I can make it faster, I can train these things better to perform better and more specific tasks?" 

Michael Segala: 07:41
Back when we were starting the company and growing the company, we latched on very quickly to Nvidia because they were pushing the realm of possibility and we said, "Oh, that looks really fun, it looks like a great opportunity to go pair to pair with them," because at the end of the day, they make GPUs. They don't solve problems, we solve problems, we don't make anything. In tandem, whoa, that's a really good marrying and we've created this real strong partnership. Today, now 60, 70-ish percent of all of our work is really in that space of what can we create from an algorithms perspective that is in and around some GPU-accelerated workload. 

Michael Segala: 08:20
A couple years ago, it was only deep learning but nowadays, you have RAPIDS and all sorts of data science stuff from a machine learning or even a graph database or all sorts of areas like that. We work on things as cool as transplants from a surgical perspective, how do you marry that up when you're thinking about it from a diagnostic and a transplant and an organ perspective, all the way to drones, all the way to autonomous vehicles and instruments, and everything in between that you can imagine because of, obviously, our expertise in the deep learning space but also this marrying, this partnership that we have with Nvidia. 

Jon Krohn: 08:57
That sounds really smart. It does sound like a beautiful marriage. Tell me about RAPID. I don't actually know what that is. 

Michael Segala: 09:04
Yeah. RAPIDS is just a software library that Nvidia put out maybe two years ago, something like that, which was their answer to only being a deep-learning-focused company, because one of their challenges was all of these organizations out there, most of them don't need the latest and greatest and most sophisticated thing that comes off the archive. That's silly. That's not real. Most of them are still using a lot of traditional statistics, math, traditional machine learning models that they said, "Well, I don't need you, Nvidia, because these are GPU workloads." 

Michael Segala: 09:38
Nvidia says, "Well, what if I accelerated the traditional," I'm going to call it, "math and ML stuff by giving you this RAPIDS platform that would accelerate all of data science?" All your Pandas, you're doing all the Pandas stuff, you're doing all the traditional ML stuff, all the graph databases, all that, it's accelerated for that stuff, which is different than just the deep learning side." 

Jon Krohn: 10:00
Wow, I learned something- 

Michael Segala: 10:03
Does that make sense? 

Jon Krohn: 10:03
Yeah, 100%. I learned something today that I need to look into immediately. 

Michael Segala: 10:06
Yeah. RAPIDS is a great platform. Like everything, it's free, it's open-source, you do your thing, you integrate it, and it's great for projects. 

Jon Krohn: 10:13
Awesome. 

Jon Krohn: 10:16
This episode is brought to you by SuperDataScience, our online membership platform for learning data science at any level. Yes, the platform is called SuperDataScience. It's the namesake of this very podcast. In the platform, you'll discover over 2,500 video tutorials, more than 200 hours of content, and 30-plus courses, with new courses being added on average once per month. All of that and more, you get as part of your membership at SuperDataScience. Don't hold off, sign up today at www.superdatascience.com. Secure your membership and take your data science skills to the next level. 

Jon Krohn: 10:54
Are you able to delve into any particular projects that you've been working with Nvidia on? 

Michael Segala: 11:00
Sure. It's not that we necessarily work with them on it. We go to the customer and the customer will buy GPUs that makes Nvidia happen and then they pay us for our services, but we go there as a strong team. It's a storytelling team. Some of the work that we're doing really spans across industry verticals. We can start, I guess, talking from a healthcare perspective. Historically, we've always worked quite a bit in healthcare just because, again, we came out of STEM so we were all physicists and working in this space in general. We have a very ... It's an easier talk track. When you're talking to researchers or talking to folks in healthcare, they really get the dynamics of the problem and you can really map what they're trying to accomplish back to what you're trying to accomplish from a data science perspective. 

Michael Segala: 11:48
Some of the interesting work that we're doing these days is really thinking about how do you use novel applications, again, thinking about data science from a novel perspective, to solve health related challenges. One of the big areas that we're working on nowadays is this concept that I mentioned it very quickly, organ transplant. A lot of our customers are these big hospitals, these big research hospitals that about once every two weeks to once every month have ... and a lot of those in the children's hospital will have children that need a total heart replacements. 

Michael Segala: 12:23
Unfortunately, they're very sick, they need somebody, a surgeon, to go in and basically replace their heart with a heart that has now recently been expired and needs to be transplanted. I'm not just talking about this use case in general, but medicine, you would think is extremely scientific and people really know what they're doing and there's this science to it. But in fact, it's really archaic in the way that they approach it and it's definitely much more of an art than it is a science. A lot of our work is trying to transition from art to science. 

Michael Segala: 12:58
This is a really good example of that where about 30% of all organs these days go wasted because they don't know how to properly match organ from recently expired individual to organ of somebody who's alive and needs it. They don't know how to match that appropriately because they don't know if this heart and the heart cavity would actually match what this heart over here needs to be. Historically, they've only used things like body weight. Is your body weight roughly similar to this other one, and if it is, sure but that is like ... We're talking about mean plus standard deviation dictating will you get a heart transplant? That's crazy. 

Michael Segala: 13:39
A lot of the work we're doing now is to think about this in a much more scientific manner of saying, "Well, I know I can take ultrasounds, I know I can take MRIs, I can take CTS where I can very, very accurately segment the entire thorax cavity and measure precisely what is that total cardiac volume available within this patient such that I know amongst all the list of applicable organs which one would be optimal and who to place into which child," or something like that. 

Michael Segala: 14:07
This whole dynamic nature of shifting it from just, "Oh, they weighed 30 kilos" to, "Well, no, I'm now going to pass this through an expertly trained system to dynamically tell you who to use this organ on should, and hopefully will, and does reduce that waste, literally organ waste, down from 30%, we're hoping, in the single digits which is phenomenal when you think about the outcomes." That's an example. 

Jon Krohn: 14:32
Yeah. That is a beautiful example. What was the percentage? I missed it. What is the percentage of organs that are wasted today? 

Michael Segala: 14:38
Something about 30%. It's in that arena. 

Jon Krohn: 14:41
30%? 

Michael Segala: 14:41
Yeah, it's a lot. It's more than it should be. 

Jon Krohn: 14:45
Yeah. That is incredible to think of all the lives that could be saved. I also wonder, this is just me thinking off the cuff, but in addition to just making use of all the organs, presumably by doing a more sophisticated analysis than just wait by doing the scanning that you're describing and presumably by getting a better fit, the outcomes are better for the patients that get the transplants anyway, right? 

Michael Segala: 15:09
100%. It's not just about here's the cavity, here's the organ. You could take that huge steps further in real presurgical planning where now, you're also mapping the blood flow. Now, you're also thinking about it from a computational fluid dynamics perspective. What is the optimal ... Once you have the heart, you're going to trim it, you're going to reshape it. It's not like you just shove this thing into the next body and you staple it and you say, "Go." It's a lot harder than that. It's all about how do you optimize the arteries, the capillaries, whatever, to match up correctly for blood flow and things like that. 

Michael Segala: 15:40
The problem just exponentially gets harder, but you can start being smarter and smarter about all steps such that you can ensure that the thing not just works but will work for more than a couple years. That's really the progression of sophistication that you want your data science to eventually get to and to be able to solve real large, big problems like this. This is just one example of thousands in the healthcare space. This idea of really taking a very manual process and making it smarter, these decision support systems, that's where the huge influence of data science is going in a lot of these industries. 

Jon Krohn: 16:18
It's exciting to hear sometimes ... I don't know where I get this inkling from. Sometimes I'm like all of the easy opportunities in machine learning have all been taken already, but having a conversation like this helps me realize that there's a crazy amount of opportunity that we're not going to need ... Yeah. 

Michael Segala: 16:34
Yeah, you're scratching the surface. I think there's a misconception and I think this is an important point to talk about. There's a huge misconception between what is solvable in an academic sense from an algorithms perspective, the Kaggle Competitions of the world, and what is practical in a real-world scenario. Just because I gave you 200 perfectly annotated CT images and you downloaded them as your train and test split on Kaggle and you can run your billion parameter segmentation algorithm, sure, but that's not how the real world works. The real work has implications that are so far beyond that where your algorithm is 1% of the total solution, if that. 

Michael Segala: 17:19
The applications of ML are so much more complex than just saying, "Here is an idealized training environment and a black box with perfect data." That doesn't exist in most scenarios. There are some companies where it is that easy, but those are really far and few in between. Most real-world data science problems in a true business context are never that trivial. You can make them seem easy but, in reality, they're really not. I think that's where a lot of these big companies fail time and time again and that's why there's always this pushback of is there really ROI to be achieved here? That's the big problem that we see in the space because it's really that misunderstanding of this low-hanging fruit problem matched with reality of outcomes. 

Jon Krohn: 18:03
That's beautifully stated. It is a recurring theme that we've had on the podcast recently not only with guests but in my FiveMinuteFriday episodes where I talk about a topic or answer questions that I've had from audience members. People working in data science have concerns about how future-proof is this career, is AutoML going to take over, and it's because of exactly these things. Yes, maybe AutoML could win the Kaggle Competition but it's not going to help understand the practicalities of what's going on with the data and make a real-world application. 

Michael Segala: 18:42
Yeah. I have a very strong opinion in that space and, again, I'm biased. Everybody has their own biases but- 

Jon Krohn: 18:48
You're biased? 

Michael Segala: 18:49
I'm biased. 

Jon Krohn: 18:50
No way. 

Michael Segala: 18:50
Of course, I'm biased. AutoML is ... There's this analogy. I gave you a calculator. Everybody, you were in grad school, I was in grad school, we all had these beautiful TI-89s. They solve complex integrals, you put all sorts of crazy stuff in that. By no means did that help me become a mathematician. These are tools in your toolbox. They made me multiply faster rather than me doing that on pen and paper, but they are just levers of tools to help make processes a little bit quicker in computation. 

Michael Segala: 19:23
Same thing with AutoML. At the end of the day, that's not going to solve a novel challenge or think about what are the implications of such challenges. For teams that are sophisticated and can use that as a lever to just speed up some of their computation all day long, awesome. But again, that is the smallest sliver of the real problem statement. They are great in context but out of context, they're actually extremely dangerous. 

Jon Krohn: 19:48
Beautifully said. 

Michael Segala: 19:49
Again, biased. I understand I'm biased, but that's at least the world that I see and we live in. 

Jon Krohn: 19:56
You and I, we can use confirmation bias together to feel very comfortable with drawing these conclusions. 

Michael Segala: 20:00
That's right. Of course. Absolutely. 

Jon Krohn: 20:04
That work dealing with the healthcare system, that sounds like it's probably public sector work to some extent? 

Michael Segala: 20:11
No, that's all private sector. 

Jon Krohn: 20:12
Oh, it's all private sector. 

Michael Segala: 20:13
That's hospital systems. 

Jon Krohn: 20:15
Oh, nice. 

Michael Segala: 20:15
That's some private hospital systems. Yeah. 

Jon Krohn: 20:17
I come from Canada so I think about that as a public sector. 

Michael Segala: 20:19
Yeah, [inaudible 00:20:19]. I understand. Yes, yes, we're a little different down here. 

Jon Krohn: 20:24
We'll use that to segue into the public sector anyway. 

Michael Segala: 20:27
Sure. 

Jon Krohn: 20:28
You have done a lot of public sector work. Do you want to tell us about some use cases that have happened in the last couple of years in the public sector and how working in the public sector is different from working in the private sector? 

Michael Segala: 20:39
Yeah, sure, of course. Yeah. About two or three years ago, we started really gaining interest and attraction and moving forward into the public sector. For folks who are listening and they don't really know the difference between public and private sector, private sector are all your for-profit companies. My company, Googles, Apples, everybody of the world who's making money by selling you stuff, that's private sector. Public sector is all the government institutions that we know about. Army, Navy, NGA, NIH, CDC, Smithsonian, all of these folks are public sector. They're all government-run organizations. 

Michael Segala: 21:23
The public sector, we believe, because we would like to believe from a feel good perspective, are super advanced because we watch these TV shows and we see NCIS, they solve a crime like that and it's beautiful, cinematic representations of nonsense. That's not the way that the world really works because the public sector, not that they aren't phenomenal at what they do, but from a technology perspective, an adoption of technology, they are laggards beyond anything that you would want to believe and it's really the challenge. 

Michael Segala: 21:55
This is where this whole AI battle between China and the US and other countries become really, really difficult is because our public sector is not nimble, not nearly nimble enough to keep up with the changing tech because they think about programs in the matter of decades, not the matter of months, and tech moves at this pace in a matter of months. So long-winded way to set the stage, what we wanted and what we are doing now is trying to be a bit of a change agent from the outside from a commercial standpoint and start working with some of these more advanced organizations within I'm just calling it DOD, Department of Defense or whomever, that is really looking to adopt some of these new core technologies. 

Michael Segala: 22:35
You might have heard of the JAIC, this centralized unit around AI for the DOD or things like AFWERX or whomever, it doesn't really matter. A lot of our work has been adopting a lot of our commercial use cases on the private sector into the public sector. That's the broad story. For them, a lot of things are mission-critical, meaning what can I do to affect a better outcome from a mission perspective. This could be geospatial if you're thinking about it from a satellite perspective. This could be things on an air force or an aircraft carrier or a plane or things as simple as supply chain, how do I get stuff from A to B. Very, very similar from the commercial field but now mapped into the public sector. That makes sense so far? 

Jon Krohn: 23:22
Unbelievably clear. You are an outstanding framer of ... You do a great job of taking a step back and framing why this is important and on top of that, your ability to reel off specific examples without taking a pause to even think about it is pretty impressive. Yeah, yeah. 

Michael Segala: 23:43
Jon, because I have to say this 30 times a day, it's just practice. Let's not fool ourselves. 

Jon Krohn: 23:48
Very good. 

Michael Segala: 23:49
In all seriousness, so a lot of the work that we're doing, some of the work that we're doing, I think, of interest to probably this group is one of the big areas that I'm most passionate about which we're thinking about from the DOD perspective is how do we do things at the edge, meaning when I'm in an environment where I don't have my laptop connected to my WiFi or I don't have access to Amazon's P3 instances and I can throw up a whole bunch of computes and sit around it for two weeks and wait for results. 

Michael Segala: 24:18
The edge really means I have a autonomous piece of equipment, this could be a drone, this could be anything, this could be a vehicle, and I want to do AI at the edge on device. Tesla's done this fantastically. I think they are probably, by far, leaders in all of the AI space. You might not like them, and I don't care about any of their politics there, but I'm talking about the fundamental technology of saying, "I'm going to embed AI on a chip in a car, disconnect it from the world, go drive." It's phenomenal. 

Michael Segala: 24:50
Think about that same application back into the government. We have drones, we have unmanned vehicles, this could be underwater, this could be on sea or wherever, this could be anywhere. How do we think about doing AI at the edge to solve novel problems across data spaces? Meaning there's a visual component, I want to look at what I'm seeing from a target perspective. Here's a person, here's a tank, here's an agriculture field, and I want to start mixing that in real-time with other spectral information. This could be electronic warfare, this could be thermal patterns, this could be RF sensing patterns, and the list goes on and on, but how do you do that in a highly congested way at the edge and do AI there? That's cool. That's amazing. 

Michael Segala: 25:34
That's really where, especially in these DOD-specific applications, that's where they're seeing the technology have to be to make them viable and reliable in the years to come. If folks are looking to get into that space in general, there's a fantastic amount of work to be had there. Again, they are just literally scratching the surface so far. There's a ton of opportunity from a career perspective, getting in and working on problems like that, which I believe are the next huge domain of real interesting topics. 

Jon Krohn: 26:04
Super cool. That is a great suggestion for people to get involved in public sector data science. I don't hear people expressing interest in that enough, and maybe it's because I'm in New York. 

Michael Segala: 26:15
I'll tell you why. Sorry, I don't mean to cut you. The problem is our government throws around money like it's nothing. Trillion dollars here, $20 trillion here, but they don't pay people well, because they have some very clear schedules called the GSA schedule where folks like yourselves who ... yourself, myself, a lot of your listeners who are really, really well academically trained, they have great credentials, they do all this great work, they can't make nearly amount of money that you can going to work with some of these big technology companies. 

Michael Segala: 26:49
You've created this very large divide where, because of capital reasons alone out of their control, they're not going to pay the same amount as you do in the commercial space, so it really depletes the talent pool and pulls people away. That's the big challenge that we really face, is just that and the bureaucracy of that. That's probably why you don't hear much people talking about it because they're like, Well, screw it, I'll go make 3X at Google." That becomes a problem. That's why. That's the same reason we have a lot of people leave academics. Nobody wants to stay and be a 60K postdoc forever when you can go make 300K at Google. It's hard. It's a financial conversation. 

Jon Krohn: 27:25
Right. On that note, though, on the note of those are very practical reasons as to why people don't go, for whatever reason, the people that I have met that do do work for the Army, I spent time in Fort Belvoir in Virginia, and those folks were great and maybe that's related to passion and resources. The demographic was different. I was teaching deep learning to them. Typically, I see a particular demographic recurringly when I teach deep learning at conferences or universities or data science academies, and it was so interesting to have people who were a lot older. In some cases, there were a large number of younger people but there were people that had been working on neural networks for 30 years. 

Michael Segala: 28:21
30. Easily. Yup. 

Jon Krohn: 28:24
Those people know a lot and they really know the detail. Sometimes I'm talking about high level TensorFlow, Keras, method calls and they're like, "So is this recommended?" 

Michael Segala: 28:32
Yeah, let's talk about the math. 

Jon Krohn: 28:35
Yeah, and I'm like, "I'm going to have to get back to you on that question." 

Michael Segala: 28:38
Yeah, of course. They want the details. 

Jon Krohn: 28:40
Yeah, exactly. 

Michael Segala: 28:40
Yeah. That's the beauty of it. The folks that are in that space, they're there because they truly care and they're extremely passionate. They're deep, deep, deep experts, especially in their technology space. Yeah, they don't care about here, this is the Keras abstraction, just make this call, they want to know the guts and that's what they're excited for and why people that do get into that space, they're lifers. They're there for 30 years easily because they're really passionate about that. If I were somebody looking for a job in this space and making everything equally, I would definitely encourage that strongly because that's where you're going to find some really, really interesting, cool stuff, granted you're going to have to want that. Yeah, that's a decision everybody can make for themselves. 

Jon Krohn: 29:27
I suspect there are also some scenarios where even though you wouldn't be bringing home a huge paycheck personally, you might have access to enormous resources for particular projects and decades' long projects like you're saying. Maybe a little bit slower moving to get going but then feeling confident about having the resources for many years to come. 

Michael Segala: 29:46
Yeah. That's the beauty of it. It is job security for life, basically. I'm sorry, I don't want to paint a picture of like they pay you peanuts. That's not realistic, but I'm saying there's a inflation in the market that has been set, especially by Silicon Valley, that nobody can really match. That's what I'm talking about. 

Jon Krohn: 30:07
I understand. All right. That is the public sector and that is super interesting. We've talked about private sector work that you've done, public sector work that you've done. What is your work like day-to-day? You obviously have a clear technical background, you have a PhD in Physics from Brown, you've done a lot of technical work, you can speak about this stuff, about any of the topics we've been covering at a reasonably technical level as far as I can tell. What is your life like day-to-day as a CEO of a scientific consulting AI company? 

Michael Segala: 30:46
Fun, hard, dynamic, different. Every day is crazy. I don't know. I don't have a good answer for you because in my role ... I'll break it down a little to give you guys a little bit more understanding. I can draw a line down the center of our company, meaning the front of that line is all about how do we get projects in the door, how do we create relationships, how do we build an organization. That's my responsibility. In the backside of the company is my two other co-founders, Dan and Mike who I did PhDs with, they have the fun side. They get to do all the execution of the work. 

Michael Segala: 31:29
This is an imaginary line because this is very crossed. My day-to-day is very clear how do we position ourselves as thought leaders, as experts in the space, and go to market with that to try to find new customers, to try to find new opportunities, to put out thought leadership, to come on podcasts, to talk to people, to run an organization. That's what I'm doing every day. 

Michael Segala: 31:52
Why that's actually really interesting, especially somebody from a scientific background, is it's not sales as in you think of sales because my selling is I'm having extremely deep technical conversations with world-class experts on our client side where it's not like we're just talking, "Oh, yeah, let's just build this model and press play," sure, but let's get to the meat of really what we're doing and stay involved as that thought leader and best-in-class perspective. 

Michael Segala: 32:20
Every day, I am talking to clients probably several times a day from literally every industry. I could have five back to back hauls from automotive, to agriculture, to the Army, to healthcare, like boom, boom, boom, boom, boom, top person of R&D, innovation, marketing, it's so varied. Then I'm reading two archive papers on the latest deep learning methods and then I'm writing SOWs on solving, I don't know, total cardiac volume estimation. It's everything, and then you do an all hands and then you go home and play with your kids. 

Michael Segala: 32:53
It's really dynamic but it's that dynamicness which is very appealing about keeping you extremely up on all things that are happening in the tech space because it's my responsibility to know that and bring that to our clients while my other guys get to have fun and actually do all the work. Does that answer your question or does that paint a ridiculously confusing picture? 

Jon Krohn: 33:12
No, no, it answers my question perfectly. It also gives me a sense that your job seems pretty good. Maybe not as fun as Mike and Dan, was it? 

Michael Segala: 33:20
Yeah, they have a fun stuff. 

Jon Krohn: 33:22
Yeah, your job sounds pretty good. I know that you have a lot of positions open right now across the board. Any data professionals, you've got an opening for them and would love to hear from people with any data skillset. What are you looking for in the people that you hire? 

Michael Segala: 33:44
That's a great question. For us, we are a consulting organization which is very, very different than a product organization. For me, we look for people that can pretty much wear multiple hats, meaning they really understand the technology and how to solve problems. 9 times out of 10, even when the client thinks that this is the problem and this is data, it's really not the case. You have to be really, really thoughtful about ways to solve hard and complex problems that you might only have an hour and no background to think about and, all of a sudden, you have to be that expert. 

Michael Segala: 34:24
It takes a lot of rigorous thought around algorithms and data and infrastructure because everybody has something extremely different, regardless of who they are, just depending on what we're trying to solve. You also have to be very curious and wanting to build the business case around that because you're not going to just hire us to sit in a room and type on our keyboards all day long and then go home. Our job is to do that but then deliver that back to the client such that they can actually do something with it from a commercialization or monetization. They have to do something with it. They have to justify paying us, obviously. 

Michael Segala: 35:00
It really is that extremely soft plus hard skills where that soft skill, by and large, it's probably the most important, to be honest with you. How do you convey thoughts? How do you convey outcomes? How do you convey statistical measures to somebody who frankly doesn't really know and might not care because that's not what their motivations on the other side of the house is? It takes a lot of thought and just mindshare around really what does it mean to deliver AI projects into the customer. It's really from a consulting perspective. Sorry, I didn't mean to drone on about all that, but somebody that wants to be client-facing but also do all the hard work and write the code and then deliver that, that's what we look for in our consultants.
 
Jon Krohn: 35:45
That response makes perfect sense. Certainly no droning. I do want to ask a little bit. You mentioned the soft skills and, obviously, that is very important in consulting. Is there some kind of unifying technology that you look for across your hires? What I mean is are there particular things? Programming in Python is probably something that's very common. It sounds like experience with deep learning might be or machine learning, it seems like almost certainly, you'd be looking for that in people that you hire. 

Michael Segala: 36:18
Of course. Both of those have to be for certain. I think about the world of data science really is three problems that you can never solve. Really, all that exists. Vision problems, these could be images, these could be video, this could be 2D, this could be 3D, it doesn't matter. Your ability to solve image-based problems, your ability to solve natural language problems, and your ability to solve signal processing problems. Signal processing is anything from financial forecasting to edge-based IoT and all the things in between. 

Michael Segala: 36:51
What we look for people who not just has done one ML problem on Kaggle, sure, that's great but that's a great start. We look for people who are a little bit more generalist across those data modalities but really understand how to go deep across all three of those, and not just doing it in a quick academic sense but what does it really mean to deploy things like that. What does that mean on Amazon? What does that mean from a GPU perspective? What does it really mean to take this and then put it on an environment from our customer that we might not know or control? It really is this broader, true fundamental knowledge of the data science space. Think about data ingestion, modeling, data output or model output. Knowing across those three big buckets is what we look for. 

Jon Krohn: 37:44
This, again, is a recurring theme with guests recently on the episode is that a lot of people at top organizations like yours are looking for data scientists who can extend, at least to some extent, up and down the stack, having awareness of implementation in edge devices in your case, cloud computing and so on. 

Michael Segala: 38:08
Yeah. We'll hire specialists. We have a team that specializes in the engineering side. We have some people we bring out who specialize just in computer vision. At the end of the day, why I said we're different than a product, it's not like you're going to spend the next three years developing just computer vision problems for the histopathology, and that's all you're going to do. That's not it. Our projects turn over very quickly so you have to want to have a general sense because you're going to work on a lot of different problems. We might have specialist but everybody at the end of the day will be well-rounded just by entropy of projects. 

Jon Krohn: 38:48
Nice. That makes perfect sense. Are there particular skills ... We're talking about what you're looking for today. Do you think that there are particular niches that people should be learning more about today so that they're prepared for the coming years of data science? I would suspect in your position, you get a really good sense of what's coming. 

Michael Segala: 39:09
Yeah. A lot of organizations are still very, very much at the POC stage, meaning here's some data that they've collected, who knows how long it's taking them to collect them, show me the technical feasibility of this use case, which is fine. Give me the data, give me a laptop, I'll sit down and I'll do the work. I'll code some stuff. I'll show you accuracies. I'll make you all these fancy ROC curves and things like that. I think that's where 90% of data science and data science teams lives today is in that space. 

Michael Segala: 39:46
I'm not talking about the Googles and the Facebook's of the world. Throw those folks away. Talking about building real companies. Not that they're not real, those guys are way too sophisticated. Those are the outliers. Those are the anomalies. I'm talking about the bulk of the real world. If you want to take the next real big step from there and say, "Where can I position myself as a real leader in the space or somebody who has almost guaranteed security in what I'm going to be doing," I would say, "Well, think about it now. Once we've proven that technical feasibility out, we need to move these algorithms into a space in production." 

Michael Segala: 40:23
What does the inferencing implications need to look like? Thinking about it from that context. Assuming that I can show you technical feasibility, what does it mean to inference against this algorithm with net new streams of data that will drift as my model changes, as my environment changes, as all the complexities really start to take charge? Think about it there. This is not just trivial ... Not that MLOps is trivial, but I'm not trivializing it by just saying, "Oh, we're just going to watch it and hope they drift." 

Michael Segala: 40:52
No, no, no. About serious thought power around how am I going to deploy these algorithms in a complicated environment, where I have to be inferenced first in my perspective and working backwards there. That's where I would focus. That could be at the edge, that could be on CPUs or GPUs, it doesn't really matter. It's still that mindshare around that from the outcomes perspective and integrating it into your final product or wherever that's going to sit. That's where I would focus and spend a lot of my time upskilling because very, very little people are there currently. 

Jon Krohn: 41:24
Beautiful. That is such clear advice for our listeners, and I think spot-on, 100%. All right. Let's talk a little bit about your journey to where you are today. We've talked about where you are and you've talked about how where you are is helpful for seeing what's happening in the future. Let's talk a little bit about your journey. You did a PhD at Brown in Physics. It sounds like your co-founders were doing PhDs at the same time. 

Michael Segala: 41:50
Yes. 

Jon Krohn: 41:53
What was your journey from finishing the academic work? Maybe even even while you were doing your academic journey a bit, was there a sense that you wanted to be doing this data science consulting then? 

Michael Segala: 42:05
Yeah. Everything from early on in my career led us to this point, amongst all, myself and our co-founders. We were extremely fortunate to be ... Everything's about timing in this world. Timing is everything. I had the wonderful fortune to be part of Brown's high energy particle physics group that was part of the Large Hadron Collider at CERN, so CERN, the largest particle collider, blah, blah, blah. I was part of a group who I had a professor, Meenakshi Narain, who's brilliant in this space. Basically, we were doing data science, this is like 2008 to 2012, at a particle collider. 

Michael Segala: 42:48
Before data science took off in the commercial space, we were doing it from an academic standpoint. We're talking about petabyte scale data collection and machine learning, it's before deep learning, machine learning algorithms on these extremely complicated, rare signal events, building algorithms, doing analysis, doing all this stuff for years and years and years which then, right when we were graduating, led to the discovery of the Higgs boson, which won a Nobel Prize. I got to be a part of that team. I was on the analysis team. 

Jon Krohn: 43:18
Wow. 

Michael Segala: 43:18
My thesis was actually on the analysis that won the Nobel Prize for the theoreticians and things like that. Listen, I didn't do anything. Everybody else, the world, the community did everything, I just happened to be very lucky. My professor said, "Hey, Mike, what do you want to do?" I said, "I want to do that," and that was a good thing. Same story with my other co-founders as well. 

Michael Segala: 43:41
Coming out of school, we had this real understanding, not full understanding, but starting this understanding of, "Wow, this data science space was really starting to take off." This is 2012. We said, "Okay, that was cool." We like that. We all went and worked, and worked for a couple years. We're very, very close still as we got out of school. We said, "Okay, we want to really take our knowledge and our passion for solving hard data and hard algorithms problem," I didn't say big data, I said hard data problems and hard algorithms problems, "into the commercial space," where all these companies will have all these problems and challenges and we could be like, "Oh, we're so smart, hire us." That's the naïve approach. "Okay so let's try that," and we did it and it worked. 

Michael Segala: 44:29
Our passion towards some of the problems that we were doing from day one at PhD materialized into now what we do every day for our clients. Back then, we were doing it to solve fundamental physics problems, but now we're doing it to do things like better drug discovery or drone applications or what is my crop yield of my agriculture farms and things like that. It's the same problem. It's the same solution. It's just the abstraction of what you're calling it has changed. 

Michael Segala: 44:57
What we do today to find better cures in medicine is literally the same thing that we do to find rare signatures of particle decays and particle accelerators. It's just all the abstraction of the logic that humans put on it, not the technology, not the data. That's been the journey. That's what we're passionate about and why it's fun, because we get to really pick that journey and stay on top of all these fun problems. 

Jon Krohn: 45:21
Beautiful. Are there any particular people that influenced you or continue to influence you today, people working in the data science space or maybe entrepreneurs or something that you look up to, and think, "This is a really great example of where I'd like to be 20 years from now," or something like that? 

Michael Segala: 45:42
That's a good question. My biggest influencers, not from a entrepreneurial space, I guess, was definitely my two advisors, both in undergrad and grad school. They were incredible, brilliant professors. They really set that cornerstone of, I think, how I envision what real rigor and problem-solving looks like and then I hope to encapsulate that myself and in the company and my employees and such, so by far them. When I think about it from who do we aspire, who do I aspire from this side of the house, up until like three weeks ago, I would have said nobody because I personally am not somebody who's driven by looking at other people's successes as a motivation. I just never was ... It's just not me personally. It's just not what drives me. 

Michael Segala: 46:33
Recently, I read Bob Iger's book. I read. Listen, I don't read anything. I listened to his book. Bob Iger is now 20 years CEO at Disney. The story that he told about the innovation that he brought to a company like Disney where, 20 years ago basically, was on the brink of disaster, they were losing money hand over fist from all their parks, and their movies, they were all flopping, and he was this huge change agent that came in and said, "Guys, listen, we have to innovate or die," and risking billions and billions and billions of dollars. 

Michael Segala: 47:08
Him alone was like, "I'm going to put the stake in the ground. I'm going to do this by acquiring the Pixars, the Marvels, the Lucasfilms of the world where everybody else says no." He's like, "No, I have a vision that is so forward-thinking," was just incredible, that story, and now you see it today, 20 years later. It's like, "Oh, yeah, no shit, buy Marvel." 20 years ago, that wasn't as obvious. To me, that was really inspirational, just like, "Wow." That's one person with such huge effects on what we all now do and love and it seems so obvious. That was a big one. I would, I guess, point to Bob Iger. 

Jon Krohn: 47:44
Beautiful. Great answer. I love that. With your perspective, doing this cutting-edge scientific consulting, seeing so many different pieces of the puzzle across the public sector or the private sector and so on, I have a question that I haven't asked a guest on the podcast yet but I've been wanting to ask, which is something that I think about a lot. Thanks to ever cheaper data storage, exponentially cheaper data storage every year, exponentially cheaper compute, firms like Nvidia allowing us to do compute at a huge scale cheaply. 

Jon Krohn: 48:23
Evermore abundant sensors, especially with 5G coming out, that's going to accelerate even more in the coming years. Obviously, unparalleled interconnectivity and data modeling innovations. You mentioned archive, people sharing things in archive at conferences at light speed with each other. The technology and machine learning and data science advances exponentially every year on all of these fronts. Is there a particular vision of the future that you get excited about for you in the coming decades or maybe for your kids? 

Michael Segala: 48:58
Yeah. Oh, my kids, it's different. Vision of the future. I would love to live in a world ... To your point exactly, I have a slide that actually says exactly what you're saying. What are the drivers for innovation? Why invest in data science? It's usually four big things. The data sets are now larger, data sets are more diverse, technology is now more accessible, and algorithms are basically open-source and constantly evolving. Those things are all great. I think one of the real at the core where I want to see the industry move and I think the biggest barrier for real adoption is the sharing of information across a diverse population. 

Michael Segala: 49:43
As a great example, I'll give you, think about back to this chest segmentation I was talking about, this estimation of the total cardiac volume for more transplants. Currently, all we have is one hospital supplying 200 labeled patient data, 200 people in the entire population, and we have to build models about that. These other hospital systems say, "I want that." But because of all these privacies and governance and all these rules and regulations, we can't just openly discuss and share information and data because of all the red tape we put on everything. That's really why the public sector is crazy difficult, because of all the red tape. 

Michael Segala: 50:24
If we don't get to a place where we've created communities and mindshares to share information such that it's not just 200 patients, it's 200,000 patients and build models that can actually attack that generalized problem, we will, I believe, quickly plateau in our innovation because we're going to analyze all the data that's siloed in our specific company, and it's going to be like, "Now what?" Let's make faster GPUs, sure, but we don't have anything net new to learn. We only can learn it faster. It's very different. 

Michael Segala: 50:55
I would love to see the world where we start to open up the floodgates of more information sharing, especially I'm not talking about to sell you more ads and crap like that, talking about solving hard problems that will affect outcomes, health outcomes, energy outcomes, financial outcomes, things that will make the population better and save lives and extend lives. That's what I think about with my son and my future children after him, is what can we do today to make his life better. I think information and data sharing has to be at the core of that because we will make very fast, very cheap stuff but, again, that's just faster and cheaper, that's not solving the problem. 

Jon Krohn: 51:35
I've heard that or read that because of COVID, some of these issues are becoming obvious to people who might possibly be able to make real changes. Public leaders might be able to make changes here because it sounds like you know this better than me, but things like if you're a big hospital system, then very early on in the pandemic, you had a lot of data on, "Oh, this is someone we need to intubate, this is someone who shouldn't intubate," various treatment options, whereas the smaller hospital system, people are dying because they don't know how to treat most people on the street. 

Michael Segala: 52:13
We'll see what's really interesting, and this is by no means a political statement nor do I care about anybody's politics, but from a Biden administration perspective, he theoretically could take a large stance, because right now would be a very opportune time to say, "Listen, we want to put data sharing and algorithms and AI at the cornerstone from a federal mandate perspective and open up." It is not something you do in four years or eight years. 

Michael Segala: 52:34
This is a momentum-building activity that would take a long time, but start that train moving and now, coming in fresh with all of the policies that are available in the pandemic to say, "Listen, let's start this today." There's really no better time or better activation energy than right now. It could happen. I don't know if it will, obviously. But, yeah, we'll see. But to your point, that's exactly what I'm referring to, is making those large legislative differences. 

Jon Krohn: 53:05
Nice. 

Michael Segala: 53:06
We'll see. 

Jon Krohn: 53:07
Thank you for all of these substantial data science guidance today. We have one last question that we always ask. Do you have any book recommendations? 

Michael Segala: 53:19
I told you mine, Bob Iger's book. Read it. 

Jon Krohn: 53:21
Oh. Yeah, of course. 

Michael Segala: 53:24
I don't even know what it's called. Something, just look it up. 

Jon Krohn: 53:26
That's okay. We'll put it in the show notes. We'll find Bob Iger's biography. We'll make sure with you that we get the right one. 

Michael Segala: 53:34
It's a biography/business-y mix, where he talks about his journey at Disney but you get some background and some perspectives of what he's done. I just found it very fascinating as like, "Wow, this guy's really an innovator." Yeah, that's the one I would suggest. 

Jon Krohn: 53:48
Cool. How should our listeners follow you or keep in touch with you to continue to get great advice like you provided over the podcast? 

Michael Segala: 53:56
Wow. I am the worst followable person. I don't have much of a following presence. 

Jon Krohn: 54:03
I don't know. I've seen you have a non-negligible- 

Michael Segala: 54:04
Yeah, I have a team of people that post for me. 

Jon Krohn: 54:08
Oh. 

Michael Segala: 54:08
No, no, no. I sometimes do mine. I'm obviously on LinkedIn. If you want to know about us, come to our website, which is being updated. Follow us, follow me, follow on LinkedIn, the company and things like that. I know we have a Twitter handle and things like that, but I don't think we use it. It's definitely- 

Jon Krohn: 54:29
LinkedIn. 

Michael Segala: 54:29
Less in your face, I would say LinkedIn for sure. 

Jon Krohn: 54:32
That is also a common thread across my guests. I think every single guest that I've had since I've taken over the SuperDataScience podcast as host has said LinkedIn is the best way to get in touch. 

Michael Segala: 54:43
Yeah, yeah. 

Jon Krohn: 54:44
Perfect. People should be getting in touch if they're interested in consulting work at the cutting edge. It sounds like you're doing the most exciting project out there. You have the best partners out there. You have tons of openings. When we were talking about this before the show, it was like data asterisk is- 

Michael Segala: 55:03
Yeah, I like star. 

Jon Krohn: 55:04
Exactly. 

Michael Segala: 55:05
Come on. If you like this space and you have skills in this space, we are open to conversations at all times of the day. 

Jon Krohn: 55:15
Yeah. It sounds like great work. All right, Michael, thank you so much for being on the show. I hope we have you on again in a couple of years and we'll catch up with all of the exciting new case studies and technological advancements that you've had in that time. 

Michael Segala: 55:29
Awesome. Thank you very much. Anybody, please be in touch, happy to talk about anything you're doing or opportunities that you may see or we can help with. Thank you, everyone. 

Jon Krohn: 55:38
Perfect. 

Jon Krohn: 55:44
Michael Segala is intimidatingly intelligent about such an absurd number of machine learning applications, or at least he would be intimidating if he weren't also so bloody friendly. Leveraging his rich AI consulting experience, in this episode, Michael enlightened us on how we're only scratching the surface of the opportunities to apply machine learning models in the real world with huge medical and societal benefits in the coming years, such as by predicting the ideal donor organ for a given patient, how being able to convert proof-of-concept data science models into actual real-world implementations is an almost infinitely large commercial opportunity in the coming decades. 

Jon Krohn: 56:27
How edge computing of deep learning models on vehicles like Tesla cars, surveillance drones, and submarines are revolutionizing the application of machine learning, how policy-enforced silos of data are holding back innovation across machine learning areas but particularly in medical applications, and how needless deaths during the COVID pandemic may finally change this longstanding policy, the pros and cons of working on data science projects in the private sector relative to in the public sector, and the three deep learning fields you need to have at least some expertise in to work at an elite AI consulting firm like SFL Scientific. That's machine vision, natural language processing, and signal processing. 

Jon Krohn: 57:09
As always, you can get all the show notes including the transcript for this episode, the video recording, any materials mentioned on the show, and the URL from Michael's LinkedIn profile at superdatascience.com/447. That's superdatascience.com/447. If you add Michael or me on LinkedIn, it might be a good idea to mention you were listening to the SuperDataScience podcast so that we know you're not a random salesperson. If you enjoyed this episode, I'd, of course, greatly appreciate it if you left a review on your favorite podcasting app or on YouTube. 

Jon Krohn: 57:42
I also encourage you to tag me in a post on LinkedIn or Twitter, where my Twitter handle is @JonKrohnLearns, to let me know your thoughts on this episode. I'd love to respond to your comments or questions in public and get a conversation going. All right. It's been a blast. Thank you for listening today. Looking forward to enjoying another round of the SuperDataScience podcast with you very soon. 

Show all

arrow_downward

Share on