Get Started

The Rabbit Hole Podcast

Welcome to The Rabbit Hole, the definitive developers podcast. If you are a software developer or technology leader looking to stay on top of the latest news in the software development world, or just want to learn actionable tactics to improve your day-to-day job performance, this podcast is for you.

rabbit-hole-with-tail.jpg

49. Machine learning with Tom Benham

On today’s episode we talk to Tom Benham about machine learning. Tom has worked in finance and risk services and has studied data science machine learning. Although the term seems to be used widely and frequently it can also be an area in which information is not always readily available. Tom gives us some insight into how algorithms are becoming useful in his field of work as well as some other professional examples. We chat about the positive sides of this,  such as the reduced chance of human error, as well as looking at the shortcomings of present configurations such as data biases. If you want an introduction to a vital concept and area of the contemporary world as well as a gateway to more info on the subject, we got you! Listen in to hear all about it!

Key Points From This Episode:

  • Tom’s background and how he went about learning about machine learning.
  • The links between machine learning and other data science professions.
  • Use cases that Tom finds particularly interesting at the moment.
  • The use of machine learning algorithms against money laundering.
  • The biases of machine learning and the importance of the data input.
  • Good podcasts and online resources about machine learning.
  • Natural lane processing and why it is of particular interest to Tom.
  • Some teach and learn take aways.
  • And much more!

Transcript for Episode 49. Machine learning with Tom Benham

[0:00:01.9] MN: Hello and welcome to The Rabbit Hole, the definitive developer’s podcast in fantabulous Chelsey, Manhattan. I’m your host, Michael Nunez. Our co-host today.

[0:00:09.8] DA: Dave Anderson.

[0:00:10.8] MN: Our producer.

[0:00:12.0] WJ: William Jeffries.

[0:00:13.0] MN: In today’s topic, we’ll be talking about machine learning. Machine learning is a mysterious thing that has appeared over on the last couple of years where everyone seems to be interested in it but It’s very difficult to learn more about it.

[0:00:25.8] DA: It’s everywhere man, everything is machine learning now. Google is machine learning and you know.

[0:00:32.8] MN: Netflix.

[0:00:32.9] DA: It’s in your phone, it’s in the air, all around us. Driving our cars.

[0:00:37.9] MN: Exactly. Constantly learning all the things about us. Machine learning, learning about humans, very strange to me.

[0:00:44.7] DA: Now we need to learn about it.

[0:00:45.8] MN: Exactly. We have a guest today. I’d like to introduce Tom Benham. Thanks for coming on down Tom, how are you?

[0:00:53.1] TB: I’m great, thanks for having me, I’m really excited to be here.

[0:00:55.9] MN: Tom, could you give us a little bit of information about yourself?

[0:00:59.2] TB: Sure, professionally. My background is more in financial services and risk and finance but a few years ago, I decided I wanted to get deeper into the space of data science and machine learning. Went back to school to start studying that and now starting to re-apply that in my sort of work environment.

[0:01:19.3] MN: From finance to machine learning, back to finance. That’s pretty – I imagine there’s a lot of machine learning happening ing finance as well. I’m sure a lot of our listeners are interested in where do you learn machine learning and I think we should kind of dive in to that.

[0:01:35.8] TB: Yeah, you want me to kick that off then?

[0:01:37.0] MN: Yeah, if you have some insight on where’d you learn. How’d you go about learning on some of the machine learning information that you’re now using in your current career.

[0:01:44.6] DA: Yeah, I’m curious if you tried to learn by yourself before you started doing that, before you started –

[0:01:50.5] TB: I mean, a little bit. I guess the introduction to me was more sort of front end in terms of data visualization and how we’re looking at like a huge balance sheets on companies and trying to iterate through the different products we have there and how we value them and that became a challenge and that I try to solve for me visualization perspective and then I just got further and further.

I got more and more interested in sort of the stack behind that and from there, getting more deeply into data science machine learning. That was sort of the genesis, my interest in it.

[0:02:23.8] DA: Right, yeah, it’s kind of a rabbit hole. As you know, many things we talk about. Yeah, for me, It was a similar kind of thing, like through the visualization aspect, okay, there’s always like cool things that we can see about it.

You know, if you do like a principle components analysis or something, you can make some image really abstract into something that you can see on a piece of paper in 2D.

[0:02:47.3] TB: Yeah, for me, that’s kind of where I started. I started to dabble a little bit I suppose in the different sub topics under machine learning but I think I’m a little bit risk averse until I thought and a little bit maybe traditional and maybe I need to get back to school to study this stuff rather than just sort of –

[0:03:02.4] DA: Yeah, sure.

[0:03:03.9] TB: I think there is a very strong and reasonable argument that all the resources you need to get into machine learning data science and pretty much online and available for free at this stage. As long as you can sort of be organized and diligent, there’s a lot of material you can use.

[0:03:17.8] DA: Yeah, it is a challenge to build the curriculum and filter through all of the information that’s out there, be it like, the kind of people that are learning these things are you know, doctorates and you know, really heavy math and there’s some great resources on like Reddit and Coursera and all kinds of like online learning programs you can go through.

[0:03:41.7] TB: One of the first things, one of the professors that teach me at Columbia there was, one of the first things he said you should realize about data science is it’s like a contact sport. You can’t do this in the theory intellect, you need to roll up your sleeves and dive into the data and the algorithms that you’re using. That’s definitely something that people want to keep in mind if they want to get into that space.

[0:04:02.4] DA: That’s right, yeah, that makes a lot of sense. It’s one thing I like read about it but then like, to actually work with data and understand the pain that comes with getting something to work and tuning it.

[0:04:15.2] TB: Yeah, this may be not familiar to a lot of your listeners. I didn’t come from a computer science programming background so that was like – I’m more from the sort of math, risk side of things, that was a bad idea for me to pick up on.

[0:04:27.4] DA: You had like the statistics and all that? Under your belt?

[0:04:30.3] TB: Yeah, it was more about learning how to think about algorithms, how to go back or not go back and teach myself some Python, sort of get out to source functional. At the end of the day you need to be building stuff, and integrating and pulling in data and writing these algorithms.

You need to – at some stage or some level, understand the theory and the math behind that but like I said, it’s a contact sport so you’ve got to get in there and start building algorithms and pulling it.

[0:04:59.9] DA: Cool, you’d say that like maybe to a degree like the understanding of, the deep understanding of that math is less important than just like having intuition and kind of using it as a blunt object?

[0:05:15.6] TB: Well, it depends, I don’t want to use the word rabbit hole but I mean, you know, the math behind it is a bit of a rabbit hole. I mean, you can go as deep as you want in that space but you know, I mean, to be effective and to be conversant, you need to have a reasonable understanding of some of the fundamental math behind a lot of the algorithms.

That was a bit of a journey for me but it was more about just getting up the curb in terms of programming and then I want to be functional in that space.

[0:05:41.4] DA: Cool, yeah. One of the things I really like about talking to people about machine learning is that there are many paths to machine learning, it’s very much like multi-disciplinary area where people will come to it from computer science, people will come to it from like statistics or math or hard science, I mean, there are people who have like – come to it from like an astrology background, they’re doing like analysis of background radiation. Nope, okay, now I’m going to be a machine learning person.

[0:06:12.2] WJ: I think you mean astronomy, right?

[0:06:14.3] DA: What did I say?

[0:06:15.0] WJ: Astrology. That’s like the –

[0:06:19.1] DA: I mean, cosmology. Yes, they are not writing the horoscopes on the path training.

[0:06:26.7] WJ: Although, machine learning would be really good at that.

[0:06:28.6] MN: Probably, yeah, you can get some NLP going.

[0:06:33.5] TB: Yeah, you can kind of just make it up I think and –

[0:06:36.3] WJ: That’s what machine learning algorithms are great at.

[0:06:38.7] MN: Just making stuff up.

[0:06:40.0] TB: That’s an interesting point because I think having a use case is actually useful in the process of learning. You have some sort of motivation behind why you’re trying to use this toolset and these capabilities. For me, it was more sort of in the finance stack or space but I fear it’s cosmology or whatever it is that you’re interested in.

[0:07:00.3] DA: Right, yeah, or generating horoscopes for adversarial networks or whatever. Yeah, what are some of the use cases that you were really attracted to with machine learning and finance?

[0:07:13.9] TB: Well, the use cases that we work on now, there’s a lot of work being done in the money laundering space.

[0:07:21.2] DA: Okay, that sounds very exciting. They’re going to be a Netflix special about this.

[0:07:27.0] TB: Yeah, there’s lots of different use cases within that as well, it seems to be one of the rich areas from a financial services space where this is being applied. Whether you’re trying to understand someone’s risk in relation to that, sort of money laundering.

[0:07:39.3] DA: That’s like the risk as in the risk that this person is a bad guy?

[0:07:44.2] TB: Yeah, the model’s kind of predicting, what’s the probability that this person should be put into that high risk bucket of a money launderer, not that they would behave badly. Just that they should have a higher level of due diligence associated with them.

[0:07:58.2] DA: I see, yeah. There’s still a human aspect where you can’t trust the algorithm entirely, it’s like trust but verify to a degree.

[0:08:10.0] TB: Yeah, there’s a lot of documentation, other things would go along with that. No, it’s more and more statuses of drive this sort of front to back processes around, how you’re dealing with your customers from that particular perspective and so –

The use case there, one of the reasons for using it is it’s standardizing how people are being considered from that perspective, rather than the arbitrary decision making from human beings.

[0:08:35.6] DA: Yeah, that’s interesting. I’ve heard about another like, pretty cool use case or like I guess, not cool, a little bit controversial actually, some judges and different parts of the US are using machine learning algorithm to score the risk factor for different defendants. Yeah, see if they need to set bail for those people.

[0:09:00.1] MN: Wow, what?

[0:09:01.9] DA: Yeah.

[0:09:03.1] MN: That’s insane.

[0:09:03.9] WJ: It’s way better than having a human do it, because right now, there is a significant correlation between the time of day and whether or not a person – it’s based of whether or not the judge just ate lunch basically.

[0:09:19.2] MN: Yeah, I’ll be really cruel if I didn’t have lunch and had to, it would suck.

[0:09:23.7] DA: Yeah, also, I’ve seen just like certain judges that are just jerks. They just set bail like more frequently, that’s really the luck of the draw if you get that guy who is mean versus the guy who is a little bit nicer or if he had lunch, then you should be – The mean guy and he didn’t eat lunch then you’re just done.

[0:09:43.7] TB: I’ve read about that as well, that particular scenario. It does get to you into another topic which is relevant to all machine learning and becoming, I think increasingly relevant in the broader AI space. That those algorithms, the other issue with that particular one is that Tony is good as the data going into it.

One of the areas that you’ll touch on if you went and start to get into the data science machine learning is the whole concept of ‘bias variants trade off’. Maybe we can get into the technicalities of that, at the end of the day, model is good in many ways as the data that’s going into it.

It may be highly tuned in some sense but it has particular biases in other aspects whether that’s upon race, gender, whatever the data is that’s being fed into that.

[0:10:31.5] MN: In machine learning, the algorithm can only be as successful as the data that goes into it because then, it would be exercised to get better.

[0:10:41.9] TB: It’s one aspect of how performant the model is definitely going to be but it’s like the data that’s going in, how you’re selecting that feature set and how your – the aspects of that model coloration.

[0:10:56.0] DA: If your algorithm is 200 bias then that means that it’s being like – it’s not fitting the data well enough so you need more data, you just throw more data at the problem then you know, you’ll be fine, it’s like just run it longer, put more data in it, okay, great.

[0:11:12.1] WJ: Yeah, I think there were some high profile instances where things were kind of embarrassingly – I think there were some high profile instances where things were kind of embarrassingly broken, like Google did an image recognition project and they were using mostly white faces to train on and when someone actually used the product to scan a black person, it classified them as a monkey.

[0:11:41.9] TB: Yeah, there’s some crazy.

[0:11:43.2] DA: Wow, jeez.

[0:11:45.3] TB: If you want something a little more, a light harder than that is there’s some good examples of training a model on blueberry muffins versus Chihuahuas. Start to look at that. See the pictures online that people can find.

[0:12:01.8] DA: They do look like blueberry muffins.

[0:12:02.8] MN: To be honest, I have a hard time figuring out if it’s a blueberry.

[0:12:07.4] TB: When you see them like side by side, just thought, this feel like okay, how did I ever tell them about?

[0:12:12.1] WJ: That’s why it’s so important to have a high quality data, because you could imagine somebody taking police records and using that to train an algorithm and then you know, we already have a lot of race bias in the policing system, in the criminal justice system right now.

[0:12:28.9] DA: Yeah, exactly.

[0:12:29.8] WJ: It’s garbage in, garbage out.

[0:12:31.1] DA: Yeah, interestingly, I know google has like some public classes and like videos that – about machine learning and one of the big things that they push is like this understanding bias and machine learning in terms of like, how it affects people of privilege of different races or genders or what have you.

Because all these things are really apparent, especially with the data that they’re working with because it’s collected from a very specific set of people who have the internet and are using computers on Google all the time, it’s a little self-selecting.

[0:13:06.6] TB: One of the other very interesting developments recently was the – I’m sure you guys are aware of the Alpha Go.

[0:13:12.7] WJ: Right, yeah, the super computer neural net that beat the world’s best go player.

[0:13:19.6] TB: Right, everyone was in the AI community, 12 months ago though that that was a good 10 years away and then someone came along and beat the best player in the world. I think it was four games to one or five games to zero.

Crushed him. Then, six months later, they built a new version of that algorithm, that neural net and there’s other related search algorithms that a company that as well. What alpha go zero, that algorithm is setup to be taught based on the pure optimization and reinforcement function with no data whatsoever. Starting from scratch. And outperformed alpha go itself, beating it a hundred to zero.

[0:14:04.0] DA: Yeah.

[0:14:05.3] MN: My god.

[0:14:06.5] DA: Adversarial networks, it’s kind of crazy what they’re able to do with these deep learning things.

[0:14:13.8] TB: Yeah, this is an interesting podcast, I was talking to you guys earlier about I16Z. We call it the revenge of the algorithm so that like – the algorithm’s now can train themselves without data at all.

[0:14:25.5] MN: Yeah, Alpha Zero is going to teach itself how to make a better algorithm that will defeat alpha zero a thousand to 0 and then it’s just going to exponentially get better and better and then we’re going to be slaves to the robots.

[0:14:39.5] DA: Yeah, it’s cool. I mean, one of my favorite movies –

[0:14:41.5] TB: We’re going there, straight away.

[0:14:44.1] DA: One of my favorite movies growing up was War Games and the only way to win is not to play. That kid hacks into the computer and talks to an AI and almost starts a nuclear war.

[0:14:55.5] MN: Wow.

[0:14:57.1] DA: It was a great movie.

[0:14:57.7] TB: But I’ve started out with finish the topic of education but I think there’s a lot of resources online if people are interested. Obviously Coursera like you said but you get –

[0:15:07.2] DA: Yeah, Coursera is solid.

[0:15:07.8] TB: Audacity as well is a really good source for.

[0:15:10.4] DA: Yeah, they have both their own nano degree programs that you can check out.

[0:15:14.5] WJ: There are also some great podcasts out there. I know you really, it is a full context board and you have to be doing it but there are hours in the day when you can’t be at your computer like when you are standing on the train or whatever.

[0:15:28.7] DA: Yeah, those hours that you are not listening to The Rabbit Hole?

[0:15:32.4] MN: But you’ve gone through all of them.

[0:15:33.7] WJ: Once you finished The Rabbit Hole then you’re allowed to listen to Learning Machines 101 is a good one and the Machine Learning Guy is another one that I really like but there are a bunch like Linear Digressions, Software Engineering Daily, he has one in Machine Learning.

[0:15:52.1] DA: Yes Software Engineering Daily is pretty great.

[0:15:54.3] WJ: This Week In Machine Learning and artificial intelligence is kind of an excessively long name. I think people just call it TWIML.

[0:16:01.7] MN: What is a TWIML?

[0:16:03.5] WJ: This Week In Machine Learning.

[0:16:05.0] MN: Oh nice, TWIML.

[0:16:06.9] DA: Yeah and hello there’s a full contact sport like you can choose the level that you’re really operating with it I guess. Like you are saying, how you can go down to the nitty-gritty for the equations and how they are actually working and doing pure implementation or like you can use an off the shelf model increasingly. You can use tensor flow or cycle at learn and just take in a linear aggression model or logistic regression model and just plug in the data.

Which is not always very easy still like getting the data on the right shape and all of that is challenging but there is a lot of tools available and increasingly API tools as well. There is services for machine learning that you can use online.

[0:16:51.7] WJ: Yeah, the Watson API.

[0:16:53.3] DA: Well yeah that’s right.

[0:16:55.1] TB: Yeah there’s some good foundational datasets as well. Iris dataset is a really good one for learning sort of classification algorithms. It’s a pretty standard one, they can get that on – well you can get them anywhere online if you just type that in. Otherwise there’s a predictor around who would die on the Titanic there is a good dataset around that. They use for –

[0:17:15.3] DA: Oh wow, this is the classic Kaggle datasets.

[0:17:18.3] TB: Exactly, it’s the Kaggle datasets in order. So that’s a good place to start as well in terms of kind of start to get your hands dirty and there’s lots of example online around how people have used those to build and classify a model.

[0:17:29.4] MN: Awesome. I mean I honestly wasn’t aware of the amount of resource that there are two get datasets that would allow you to do these machine learning exercises but it’s really cool.

[0:17:38.6] TB: You know there’s many, there’s tons of index data at different universities as well.

[0:17:42.9] MN: Cool, hopefully we can definitely add them in the shownotes and have people look at it that way.

[0:17:51.0] DA: Yeah.

[0:17:51.8] WJ: Are there any cool applications in the financial space for this that we haven’t touched on yet?

[0:17:58.3] TB: Yeah, I mean one of the use cases that we are focused on right now that’s in the area that is particularly interesting to me is around the natural lane processing. So we are looking at a use case around how do institutions map the internal policies and procedures to the regulatory requirements that they need to comply with. So this is an area where this sort of document set or corpus is grown organically overtime with no standardized formats and ways of collating that information.

So we’re looking, we’ve been building some algorithms that will essentially run over that corpus versus now any external regulatory corpus to correlate between those documents and so that helps us get insight, it helps our clients get insight in terms of where they may have gaps in terms of how they are complying with particular regulations from a documentation or from an execution perspective and it is a much quicker sort of time to market in terms of assessing the risk around that particular use case. 

So I am going over pretty high level but that’s one area that we are focused on and it’s been getting a reasonable amount of traction with our clients as well.

[0:19:08.7] MN: So being able to use the machine learning to learn some of the contractual aspects of trading, is that the idea?

[0:19:19.5] TB: Well it’s not so much contract related. It’s more regulations, there is a lot of regulations yeah. So it’s really just essentially running these algorithms of this large corpus of documents and you may put in a bunch of other stuff in there to increase variability of that type is that but essentially what we are trying to do is, like I said, just correlate between their internal documents and their external documents and then we have our subject matter.

If it’s come in and sort of validate that the model is predicting well in terms of how they correlated and then once you get confidence in that then you go to the area where it says they’re not correlated. You work out where the gaps are and we can work on how to close that with our clients.

[0:19:56.7] DA: That’s cool.

[0:19:57.2] TB:  Natural lane processing is a massively interesting space to me and that is a really good use case that we go.

[0:20:02.8] DA: So that’s really taking something that instead of a computer is getting all of these documents, it would have been some poor intern like sifting through piles of documents.

[0:20:11.9] TB: Well lawyers actually. They are lawyers, not some poor intern.

[0:20:15.8] DA: Oh okay, wow high stake.

[0:20:18.2] TB: Yeah so that’s it.

[0:20:20.9] MN: Oh that’s really interesting.

[0:20:23.3] TB: But again some of the traditional risk spaces that we operate in, whether it’s credit risk, well credit risk is a good one, I mean they are using the same algorithms, the same models that people associate with machine learning and sort of newer modeling capacities but these are – one of the things that you do land I think when you are getting up the curb on some of these stuff, a lot of these models have been around for a while.

It’s just that they didn’t have data or the compute power to apply them into lots of used cases. So there’s a lot of crossover between traditional ones that are around.

[0:20:54.4] DA: It is certainly cheaper than paying a lawyer to look through all the stack.

[0:20:58.4] MN: Well yeah.

[0:21:00.5] WJ: Watch out lawyers, we’re coming for your jobs.

[0:21:05.0] DA: Lawyers, accountants.

[0:21:09.0] TB: I mean you guys would get a lot of these I think in the work that you do, when you go in to see clients and customers. I mean a lot of the work that people do is just massaging data and moving it between systems. High paid people are doing this though. So the more you can take humans out of loop of that process, the better. I mean I don’t know how much experience you have been doing that but we do a lot of that.

[0:21:34.4] WJ: We’re very used to having our jobs automated because that happens pretty much every two years like everything that we used to have to do, there is some framework now and you have to learn all new stuff.

[0:21:44.8] DA: Yeah, I feel like the one advantage of our jobs though is that we are very adaptable and used to being automated out and we love it. We love it when we don’t have to do that thing that we used to do manually.

[0:21:56.6] WJ: That’s true.

[0:21:58.2] MN: I mean until they automate our jobs.

[0:22:01.2] WJ: Until they automate the process of automating away the things that haven’t automated yet.

[0:22:05.0] DA: I would love that too, that will be great.

[0:22:08.0] MN: Oh man.

[0:22:09.4] DA: Cool, should we do any teach and learn?

[0:22:12.4] MN: Yeah, I think it’s the perfect opportunity to start some teach and learns. Anyone here have a teach and learn they want to start off with?

[0:22:19.4] WJ: I can start. I have been looking at some Ethereum block chain stuff around solidity and the Ethereum virtual machine and how that works. I was reading recently about JSON RPC which is how the Go Ethereum server that you start when you are working with Ethereum communicates with an individual node, how you can communicate with an individual node and it’s this protocol called JSON RPC which is an implementation of the remote procedure call or RPC protocol.

And the way it works is if you want to execute a command on the node, you can use this protocol which is very similar to Rust and it allows you to send the JSON version. You send JSON back and forth and each JSON object has a signature. Something like the version of the protocol, the name of the method that you want to call, any parameters that you want to send to it and ID so you can match a call with a response.

[0:23:24.7] MN: I see. So just like is that the basic way of getting any Ethereum smart contracts out into the node pretty much or?

[0:23:34.2] TB: I thought Ethereum was a crypto currency.

[0:23:37.6] MN: Oh it is.

[0:23:38.3] WJ: Yeah, it’s a crypto currency but the thing it makes it valuable is not scarcity the way it is for Bitcoin because there is no cap like there is on Bitcoin. Bitcoin there will only ever be 21 million coins. With Ethereum, you can continue to make it. The value comes from the ability to write smart contracts and those contracts then live inside of the block chain which means that everybody has a copy and you can enforce them easily. Nobody can – it’s harder to argue about.

[0:24:13.0] TB: It’s absolute transparency basically on what’s been agreed

[0:24:16.7] WJ: Yeah and you can execute the contracts inside of the Ethereum virtual machine. So assuming you’ve put in currency as like for Escrow for example, you could have the contract in for the withdrawal of Escrow. So the smart contracts are why people are interested in investing in Ethereum, interested in using it. So as a New Year’s Resolution, I have to learn about this, thanks to these guys I have a smart goal and so I was researching that.

I actually am not sure if that is how you get a smart contract into the block chain. I haven’t gotten that far. I think it may not be but we’ll find out.

[0:25:05.1] MN: I think it’s a start.

[0:25:06.2] DA: I think the theme that we are going out that was like watch out lawyers we are coming for your jobs, contracting and reading documentation. Tom do you have anything?

[0:25:18.5] TB: Yeah, so we were talking, I think before we started here, so my recent thing and I don’t know if it is really going to be teaching, unless you are just learning from my abject failure but it’s just that.

[0:25:28.8] MN: I think it’s that you learned something.

[0:25:32.1] TB: Yeah, so this is in the spirit of contact sports. In the last six weeks I started taking up Jujitsu. If my wife ever listens to this which she probably won’t, she’ll cringe when I bring this up but yeah that’s my new thing is getting into Jujitsu and getting my butt kicked by -

[0:25:52.1] MN: And flipped.

[0:25:52.8] TB: Yeah, that’s happened a couple of times. Like I said, I think I almost broken my elbow twice, I may have cracked a little rib and this is over than it’s like six weeks of starting this.

[0:26:04.0] DA: You know that’s badass though.

[0:26:05.7] MN: That’s Jujitsu 101 I guess is required for you to break a rib, is that the rule?

[0:26:11.7] TB: I don’t know but it was weird because you literary started cold, six weeks ago and my first class you go in there at the end, you warm up and do a few things and then you might drill a couple of different moves and try that out with a partner and then the last part of the class is like, “Okay you got five minutes, you just roll with someone right? And you know the goal is like submission. So it’s actually better to be with a higher belt because I know what they are doing.

And they are not trying to beat you because the low belt you go, the more aggressive someone seems to be. So you learn and you start getting up the curb pretty quickly to some degree.

[0:26:50.5] DA: Man, I don’t know if I have to choose between contact sports, between that and machine learning, I will choose machine learning. It sounds tough.

[0:26:59.3] TB: It’s quite meditative because your mind is pretty much focused on what you’re doing. You are not drifting off.

[0:27:05.0] MN: Just not getting your ribs broken, I imagine that is the thing that on your mind,, not to get my ribs broken.

[0:27:09.6] TB: Yeah but I recommend it.

[0:27:13.0] DA: Nice.

[0:27:14.3] WJ: Watch out Dave, next time you show in machine learning I’m going to sneak up on you and just break a leg.

[0:27:18.1] MN: What? Geez.

[0:27:20.4] WJ: So you’ve got to come do Jujitsu with me man.

[0:27:23.8] MN: It’s going down.

[0:27:24.4] DA: Nice, cool.

[0:27:25.5] MN: So don’t get flipped either in machine learning or Jujitsu. Yeah this about wraps up the episode. Thank you Tom, Tom Benham, ladies and gentlemen. Thank you so much for coming on down and sharing some of that.

[0:27:39.5] TB: Thank you.

[0:27:40.6] MN: We got to learn about machine learning that’s my learn for today, thank you. Thank you so much for coming down.

[0:27:45.0] TB: Thank you guys, it was awesome to be here.

[0:27:47.1] MN: I’d like to thank my co-host, Dave Anderson, thank you.

[0:27:52.1] DA: Thanks man.

[0:27:52.9] MN: And our producer, William Jeffries.

[0:27:55.0] WJ: Happy to be here.

[0:27:56.1] MN: Actually before I close out the episode, Tom how can people reach out to you?

[0:28:00.4] TB: Yeah, sure I mean if people want to contact me they can just find me in LinkedIn, that’s probably the easiest way to do it. It’s Thomas Benham and I work at PWC. So those data points should be enough to navigate.

[0:28:11.8] MN: All right, we’ll let the machine learning do the rest. Awesome, I’m Michael Nunez. Feel free to hit us up at twitter.com/radiofreerabbit and if you haven’t, please give us a five star rating on iTunes and subscribe.

This is The Rabbit Hole, we’ll see you next time.

 

Links and Resources:

The Rabbit Hole on Twitter

Tom Benham on LinkedIn

Python

Alpha Go

Coursera

Learning Machines 101

Machine Learning Guide

Linear Digressions

Software Engineering Daily

This Week In Machine Learning

Watson API

Iris Data Set

Kaggle Datasets

Ethereum

Escrow

Comments