AI - Artificial Intelligence

[Jurgis]

The paper I was talking about is

"Dow Jones Trading with Deep Learning: The Unreasonable Effectiveness of Recurrent Neural Networks"

to be presented at http://insticc.org/node/TechnicalProgram/data/presentationDetails/69221

 

The paper is not publicly available, but you can ask the authors for copy. I have a copy and can send it to people interested, but I won't post it here publicly. PM me if you want a copy.

 

Couple comments on various things previously mentioned now that the paper is semi-public:

 

- The paper predicts daily close of DJIA from daily open value + opens of previous n days (2-10).

- The trading algorithm is simply buy if predicted close > open and sell otherwise. If you cannot buy (already have position), then hold. If you cannot sell (already hold cash), then hold cash.

- Authors use training data from 01/01/2000-06/30/2009 and test data from 07/01/2009 and 12/31/2017. This somewhat answers the critique that training is from bull market: it's not. Testing is not completely from bull market either.

- Authors use pretty much vanilla LSTM, so IMO the critique that "1000s of academics looking for signals and the winners publish a paper" or that they have tweaked/over-fitted the model until it worked does not seem to apply. (It's possible that they messed up somehow and used testing data in training, but they seem to be careful, so it doesn't seem very likely). This is really vanilla IMO without much tweaking at all. Which makes the results surprising BTW.

- I have some other comments, but I'd rather discuss this further with people who have read the paper, so I won't post them now.  8)

 

As I mentioned, I spent a bunch of time reimplementing what these guys presumably implemented.

I do not get their results. My results are pretty much at level of random guessing, i.e. the accuracy is around 48-52% while they get up to 80% accuracy.

It's quite possible I am not doing something the same way they did.

It's possible that their implementation or testing is messed up somehow too. But it's hard to prove that. Maybe they'll opensource their implementation sometime in the future.  8)

 

If anyone is interested to get together (online through some tools?) and go through the paper and/or my implementation, we can do it. PM me and we'll try to figure out what would work best.  8)

 

I dont know what the authors did but ill reiterate from before vanilla LSTMs do little better than guess on the stock market.  They probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set.  This is why typically papers like this are not believed anymore in the ML literature.  Try adding some stuff like attention or skip connections and whatever else is hot now (I'm not sure) and didnt someone recommend GRUs instead.  I have some other ideas you can use like Gaussian Processes to estimate realtime covariance matrices, but your better off looking at the literature first than trying out hairbrained ideas that might not work.  It's really not a trivial excerse to outperform the market with ML.

 

Ah, I think I see where there is a miscommunication between us. :)

 

My goal is not to outperform market with ML. My goal is to understand whether what is proposed in this paper works and if it does not then why.  8)

 

You are possibly completely right that what authors propose does not work.

I just want to understand how they got the results they got.

 

You've said "probably had like 1000 GPU and tested thousands of hyperparameter configurations and "overfit" the test set." before.

I don't think that's the case at all. If you read the paper - which you haven't so far - you can see that their training is really simple and there's no "thousands of hyperparameter configurations". Which is baffling in itself. I have some suspicions of what could be wrong, but it's not productive to discuss it if you just dismiss the paper offhand. Which is BTW your prerogative - if that's where you stand, that's fine and I won't bother you with this further.  8)

 

You are entirely correct that I haven't read the paper and maybe I was too hasty in dismissing the paper.  I wouldn't mind a copy of the paper if you don't mind sending me one. 

 

That being said here is my reasoning in more depth.  The authors seem like they are in ML acadamia, so I made a couple assumptions.  1.) It didnt look like their paper made it to one of the premier conferences and maybe its because they aren't big names but likely its because people have been training LSTMs on stocks for a long time and vanilla LSTMs dont work well and I think everyone in the ML community is suspicious of 80% hit rates using a vanilla LSTM on indices for good reason and they likely didn't do anything special to assume that they didn't just get "lucky" with their model.  the  reason they got "lucky" is number 2) typically papers dont discuss the hyperparameter search they go through to find the exact correct configuration, so even if they didn't say they tested 100s/1000s of hyperparameters they might have and likely did (although yes i didnt read the paper). Unless they specifically say there were few or no hyperparameters to test or they tested only a few of them, you should assume they did test many.  This is a  dirty secret in ML, you come up with a new technique and you dont stop testing hyperparameter choices the model until you get good results on both the test set and validation set.  Then you submit to to a journal saying this method did really well because it outperformed on both the validation set and test set.  But you stopped right after you get a hyperparameter choice that met those criteria which strongly bias your results upward.  This is related to p-hacking.  This is a perfectly natural, but bad thing people do and usually means most papers have performance that can't be matched when trying to reproduce them.  You can pick basically any method of the thousands that have been proposed and if it doesn't have over 1000 citations (and the method actually seems useful) this is probably one reason why. 

 

Now you maybe you are right and something else may be missing, but if I had to guess I think its a good chance the authors just got "lucky".  BTW why dont ask the authors for their code.  Its customary to either give this stuff out or post it on github. 

 

As a side note: Even a vanilla LSTM has many hyperparameters: number of states, activation type, number of variables to predict, test/train/validation breakdown, number of epochs, choice of k in k fold validation, size of batches, random seed, how they intialized weights (glorot, random nomal, variance scaling..) for each weight in the ANN, the use of pca or other methods to whiten data, momentum hyperparameter for hillclimbing, learning rate initialization, choice of optimizer...

 

My point is that even with a vanilla LSTM the author can pull more levers than can be hope to be reproduced if you don't know absolutely everything maybe even down to the version of python installed to reproduce the pseudorandom number generator.  No doubt some of these choices will be mentioned in the paper, but many of these choices won't be typically, which makes any reproduction difficult.  And typically the authors are the only ones who are incetivized to keep trying hyperparam configurations until one works. 

 

The real papers that are sucessful are typically methods where either its not impossible to get a reproducible and externally valid hyperparamter configuration, or something that is relatively robust to hyperaprameter choices.

 

I sent you the link to the paper.

If you look at table 1, there's couple things to notice:

Yeah, for Adagrad, the accuracies are all over the place. But for Momentum and Rmsprop they are all quite similar and way higher than 50% (which would be random guess). So I think this somewhat shows that they did not just pick a single lucky combination of what you call hyperparams. You can still argue that perhaps there's a lucky hyperparam that is not shown in Table 1. That's possible, but I guess it's becoming less convincing. ;)

 

OTOH, I did not run all the combinations they presented in Table 1, but from what I ran, the results were way more stable and clustered at 48-52% range. So I wonder why they are getting much wider dispersion than I do and why their results have so much better accuracy. So I wonder if their results are correct.

 

In other words, you question their results because you think they hyperparam hacked. I question their results because I think there's another issue somewhere. But I don't know what it is.

 

I think you're a bit overstating the instability of the runs. Yeah, there's definitely hyperparam hacking, but IMO - and I'm not a huge expert - the big difference comes from network architecture hacking rather than version of python, random seed, etc.

Also I think you're mostly talking about papers/work where someone tries to squeeze out couple % gain on a widely studied problem where tons of methods have been applied in the past. I'd be more inclined to agree with you if these guys were at 53% accuracy in single or couple tests. But with the number of results in 70% range, I think there's something else going on. But since I don't know what it is, your argument might be still weightier than mine. ;)

 

 

 

[cameronfen]

not going to quote again but I took a look at their paper.  You are certainly right the results seem to be robust to the param choice they show.  That being said I wonder if there was something to how they cleaned the data because 65% performance on a feed forward ANN is phenomonal.  usually LSTMs only perform like 54-56% at most with a bit of work (but the papers I've read use intraday data).  I'm curious also why they published if there model performed so well.  why not use it themselves if they get good results?

[Jurgis]

not going to quote again but I took a look at their paper.  You are certainly right the results seem to be robust to the param choice they show.  That being said I wonder if there was something to how they cleaned the data because 65% performance on a feed forward ANN is phenomonal.  usually LSTMs only perform like 54-56% at most with a bit of work (but the papers I've read use intraday data).  I'm curious also why they published if there model performed so well.  why not use it themselves if they get good results?

 

I have to compliment you: you do hit the right spots.  8)

 

I also thought that their data cleaning/preparation was suspect. So I ran on data without their "missing dates" transformation. I thought I'll get bad results with that and then the results will jump when I do their transformation, which would prove that their results are caused by bad data preparation. No cigar. I get the same bad results with original data and with data prepared based on their description.

Which does not prove that their preparation wasn't broken... it might be broken in a way that's not described in the paper. Or maybe the description does not match what's in the code.

 

Anyway, maybe this is enough time spent on this paper. Maybe the right thing to do is to wait if they gonna publish the code (or ask for the code). Or just conclude that their results are broken and we just don't know why.  8)

 

I'd still be interested to discuss with someone my implementation and where it might be different from theirs. Just to see if I missed something obvious or did something wrong. But my code is hacked up mess, so it's probably not a high ROI for anyone to look at it. 8) I could put my code on github... oh noes, I'm too ashamed of the quality to do it...  ::)

 

Anyway, thanks for discussion so far.  8)

 

 

Ah, regarding

I'm curious also why they published if there model performed so well.  why not use it themselves if they get good results?

 

I write this off as academia. People may be more interested in results/papers/thesis (I think this was master's thesis for one author) than in applying it in real life.

Almost nobody from the people I know transferred their thesis/papers into actual startup work. Maybe it's more common nowadays and in certain areas, but it's likely not very prevalent. I guess this area/paper would be easier to transfer into money making than other theses, but they still might not be interested.

A valid question though. I'm not cynical enough to suggest that they know their results are broken and that's why they published instead of using them themselves. I somewhat believe people don't consciously publish incorrect results. But who knows.  ::)

[cameronfen]

haha thanks.  Just to clarify though, I dont think academics are conciously cheating.  But when you are pushed to get positive results, you typically search until you find something, then you just don't check to make sure you didnt screw something up. 

[Liberty]

https://blog.openai.com/openai-five-benchmark-results/

 

"Yesterday, OpenAI Five won a best-of-three against a team of 99.95th percentile Dota players: Blitz, Cap, Fogged, Merlini, and MoonMeander — four of whom have played Dota professionally — in front of a live audience and 100,000 concurrent livestream viewers."

 

https://arstechnica.com/gaming/2018/08/elon-musks-dota-2-bots-spank-top-tier-humans-and-they-know-how-to-trash-talk/

 

"OpenAI is using 128,000 cores on Google's Cloud Platform. The bots learn the game from scratch: initial versions will just wander aimlessly and at random as the game plays itself out. As thousands upon thousands of games are played, it figures out which actions will improve its chance of winning."

[Jurgis]

https://www.bloomberg.com/news/articles/2018-10-09/the-big-problem-with-machine-learning-algorithms

 

The $11 million BUZZ US Sentiment Leaders ETF has returned an average of 21 percent a year from inception in 2016. It gained 25 percent this year through Oct. 5.

 

https://www.etf.com/BUZ

 

[UNF2007]

Wen't to order a pizza today from Dominoes, when I picked the delivery option they asked if I would be willing to have the pizza delivered by a self driving car, then had me mark a point on a map near my house for it to arrive to and said I would get a code to unlock the car when the pizza came. It turned out not to be available in my area, but maybe evidence they are piloting this in other areas. Has anyone else seen this? I've always thought the adoption for this stuff will be way faster then the average person suspects. Also changes the way to think about delivery, maybe restaurants that would not have wanted to provide a delivery service in the past will change, if you could contract with a third party company to provide the service?

[Jurgis]

https://www.theguardian.com/environment/2018/oct/20/space-robots-lasers-rise-robot-farmer?CMP=share_btn_tw Robo Farming in UK

[DTEJD1997]

Wen't to order a pizza today from Dominoes, when I picked the delivery option they asked if I would be willing to have the pizza delivered by a self driving car, then had me mark a point on a map near my house for it to arrive to and said I would get a code to unlock the car when the pizza came. It turned out not to be available in my area, but maybe evidence they are piloting this in other areas. Has anyone else seen this? I've always thought the adoption for this stuff will be way faster then the average person suspects. Also changes the way to think about delivery, maybe restaurants that would not have wanted to provide a delivery service in the past will change, if you could contract with a third party company to provide the service?

I have not heard of ANY self driving vehicles on American roads being used in regular commerce.  It is all R&D at this point, unless I am tremendously mistaken.

 

Are you in USA?

 

Could this perhaps be a TEST to see if consumers are willing to do this?

 

Much like automated vehicles for Uber & Lyft, I wonder if there is really a commercial market for this.

 

How much of a savings will a typical Domino's franchisee save by having self driving vehicles for pizza delivery?

 

A pizza delivery driver provides their own vehicle and makes about minimum wage from the franchisee.  Tips provide the rest of his earnings. 

 

In my area of the world, the pizza delivery driver almost always provides their own vehicle.  For an automated delivery, the franchisee is going to have to buy the vehicle.  How much will that cost?  I would guess at least 50k.  How much cheaper will it be to run than having a delivery driver?  What will the savings be if any?

 

$50K outlay for a delivery vehicle is going to be a HUGE percentage outlay for the typical franchisee.  The typical Dominos would probably also need MORE than 1 vehicle.  I would think they would need at least 2, if not 3 or 4 (more?).  I've heard rumors that the typical Domino's franchise costs as little as $150k to about $350k in investment.

 

Way back when, I used to be a delivery driver when in college.  On busy nights, which were usually Thursday, Friday and Saturday (sometimes Sunday), we would have up to about 10 drivers.  On slower weekday nights, we would have 3-4.  Drivers would also not usually work 8 hour stretches.  Drivers would typically be promised a minimum of 4 hours, and then those who wanted to leave earlier could (if business was slow).  So the franchisee could adjust labor pretty quickly.  With self driving vehicles, you've got a constant expense, can't let a robot go.

[SharperDingaan]

Look into the mining industry.

https://www.trafficsafetystore.com/blog/mining-operations-autonomous-vehicle-technology/

https://www.nbcnews.com/mach/science/robots-are-replacing-humans-world-s-mines-here-s-why-ncna831631

 

Labour is expensive, and in increasingly short supply. The technology works best in controlled spaces, and the more hostile the environment the better (self-driving ore loaders, where explosive gas/rockfall is a very real danger). More importantly it's not visible to Joe Public (hence no protests against job loss to the 'machines') and it saves lives, lungs and limbs; but it costs the unskilled their livelihoods.

 

It's also dirt cheap. 10M in equipment under a capital lease might cost 8%/year ($80,000) and replace 2-3 people (2.5) at 80-100K/yr net of benefits. Per the P&L; spend an extra 80K to save 200-250K, and collect premium savings on reduced health and safety claims as added bonus. The same mathematics that is at work in automated warehouses and airport baggage handling. Great for productivity, but if you were one of those 'displaced', you're pretty much out of a job for good.

 

The Corporate Social Responsibility (CSR) issue.

 

SD

 

 

 

 

[rkbabang]

Wen't to order a pizza today from Dominoes, when I picked the delivery option they asked if I would be willing to have the pizza delivered by a self driving car, then had me mark a point on a map near my house for it to arrive to and said I would get a code to unlock the car when the pizza came. It turned out not to be available in my area, but maybe evidence they are piloting this in other areas. Has anyone else seen this? I've always thought the adoption for this stuff will be way faster then the average person suspects. Also changes the way to think about delivery, maybe restaurants that would not have wanted to provide a delivery service in the past will change, if you could contract with a third party company to provide the service?

I have not heard of ANY self driving vehicles on American roads being used in regular commerce.  It is all R&D at this point, unless I am tremendously mistaken.

 

Are you in USA?

 

Could this perhaps be a TEST to see if consumers are willing to do this?

 

Much like automated vehicles for Uber & Lyft, I wonder if there is really a commercial market for this.

 

How much of a savings will a typical Domino's franchisee save by having self driving vehicles for pizza delivery?

 

A pizza delivery driver provides their own vehicle and makes about minimum wage from the franchisee.  Tips provide the rest of his earnings. 

 

In my area of the world, the pizza delivery driver almost always provides their own vehicle.  For an automated delivery, the franchisee is going to have to buy the vehicle.  How much will that cost?  I would guess at least 50k.  How much cheaper will it be to run than having a delivery driver?  What will the savings be if any?

 

$50K outlay for a delivery vehicle is going to be a HUGE percentage outlay for the typical franchisee.  The typical Dominos would probably also need MORE than 1 vehicle.  I would think they would need at least 2, if not 3 or 4 (more?).  I've heard rumors that the typical Domino's franchise costs as little as $150k to about $350k in investment.

 

Way back when, I used to be a delivery driver when in college.  On busy nights, which were usually Thursday, Friday and Saturday (sometimes Sunday), we would have up to about 10 drivers.  On slower weekday nights, we would have 3-4.  Drivers would also not usually work 8 hour stretches.  Drivers would typically be promised a minimum of 4 hours, and then those who wanted to leave earlier could (if business was slow).  So the franchisee could adjust labor pretty quickly.  With self driving vehicles, you've got a constant expense, can't let a robot go.

 

Delivery is a pizza place's bread and butter, so I think they might be the last to convert to driverless delivery. But there are a ton of other restaurants who do not deliver today.  What if a company say Uber, who had a large fleet of driverless cars already, offered restaurants a delivery service that they could subscribe to for less than it would cost them to hire delivery drivers and not have the hassle of having more employees to manage, and had a pricing scheme where it would be cost effective even if customers only seldom ordered for delivery?  Right now delivery is basically just pizza places or using Uber Eats and other services in which the customer pays extra for a service to go and pick up takeout for them.  Most sit down restaurants don't hire delivery drivers because the vast majority of their customers eat in the restaurant rather than order takeout.  But in the above subscription service scenario where offering delivery would be cheap and easy for any restaurant to offer, a restaurant could be at a serious disadvantage if it didn't offer delivery when almost all other places do.  My wife and I order takeout all the time from restaurants which are usually too crowded to get a table on the weekends on short notice, it would sure be nice to just have it delivered rather than need to go and pick it up.

 

[rkbabang]

Labour is expensive, and in increasingly short supply.

 

This is the largest reason that this is going to happen.  I don't know about the rest of the country, but up here in New Hampshire all everyone talks about is the lack of people to fill jobs.  There was just a new local restaurant built that hasn't opened because they can't hire enough people to work there and another one that was only open a short while and closed for the same reason.  My kids have had no problem finding jobs that pay way over minimum wage for bagging groceries or working at restaurants, anything.  We can't get our furnace cleaned for the season until January, the company told my wife that they are severely short staffed and can't find anyone to hire.  Every McDonalds in the area has one person working the register and 4 automated ordering stations.  They all remodeled this year to install them.  It is the same everywhere you go, big "hiring" signs in the windows of most businesses.  Finding contractors to do anything is close to impossible, because they are all short staffed and booked for months. I know it isn't just New Hampshire, my daughter just moved to FL a month ago, she applied for 3 jobs the first day and got all three. She took two of them and the 3rd one begged her to not refuse the job, the hiring manager kept calling her and even texting her, and even offered her more money.  She's 18 with just a high school diploma and some restaurant experience.  One article I read quoted someone who said "what are teenagers doing with their time",  I don't know, but it's great for the ones who do want to work.

 

 

https://www.wmur.com/article/portsmouth-restaurant-closes-due-to-lack-of-workers/24192393

 

https://www.sentinelsource.com/business_journal/the-labor-shortage-good-help-is-hard-to-find-in/article_740cee2c-9897-11e7-bea7-d78d275357cf.html

 

https://www.laconiadailysun.com/news/local/good-jobs-few-workers/article_3ec3fcdd-9265-5208-b4f0-5d1df7e36033.html

 

https://www.marketplace.org/2016/12/12/economy/low-unemployment-rate-new-hampshire-creates-labor-shortage

 

 

Automation, wherever it's possible do to, is going to be a necessity.  I think the pace of automation in the short to medium term is going to be surprising to many.

 

[UNF2007]

I'm in the US, Colorado right now. Your right that it may have just been consumer research disguised as a legitimate service, it crossed my mind. Another angle I was thinking about is how much of an incentive is it for the customer (if given the choice) to pick automation where no tip is involved or expected? As it stands now I have to pay a delivery charge and tip also. Maybe it will take longer to implement then I think. Just to examine it a bit more, quick google says the average domino's sells about 132 pizzas a day, I don't know what % is delivery, maybe 80%? So 105 a day, maybe allocate 25% of the volume to the automation to start, so 26, with a 5 dollar delivery charge, 130 a day? Over a week 920, a month 3680. Seems like it could pay for itself, maybe even increase profitability as you would depreciate it and could also deduct interest off a loan if it was financed? Obviously it would be a franchise by franchise decision, and your volumes would have to support it, but I bet for the right locations it could boost profitability.

 

Wen't to order a pizza today from Dominoes, when I picked the delivery option they asked if I would be willing to have the pizza delivered by a self driving car, then had me mark a point on a map near my house for it to arrive to and said I would get a code to unlock the car when the pizza came. It turned out not to be available in my area, but maybe evidence they are piloting this in other areas. Has anyone else seen this? I've always thought the adoption for this stuff will be way faster then the average person suspects. Also changes the way to think about delivery, maybe restaurants that would not have wanted to provide a delivery service in the past will change, if you could contract with a third party company to provide the service?

I have not heard of ANY self driving vehicles on American roads being used in regular commerce.  It is all R&D at this point, unless I am tremendously mistaken.

 

Are you in USA?

 

Could this perhaps be a TEST to see if consumers are willing to do this?

 

Much like automated vehicles for Uber & Lyft, I wonder if there is really a commercial market for this.

 

How much of a savings will a typical Domino's franchisee save by having self driving vehicles for pizza delivery?

 

A pizza delivery driver provides their own vehicle and makes about minimum wage from the franchisee.  Tips provide the rest of his earnings. 

 

In my area of the world, the pizza delivery driver almost always provides their own vehicle.  For an automated delivery, the franchisee is going to have to buy the vehicle.  How much will that cost?  I would guess at least 50k.  How much cheaper will it be to run than having a delivery driver?  What will the savings be if any?

 

$50K outlay for a delivery vehicle is going to be a HUGE percentage outlay for the typical franchisee.  The typical Dominos would probably also need MORE than 1 vehicle.  I would think they would need at least 2, if not 3 or 4 (more?).  I've heard rumors that the typical Domino's franchise costs as little as $150k to about $350k in investment.

 

Way back when, I used to be a delivery driver when in college.  On busy nights, which were usually Thursday, Friday and Saturday (sometimes Sunday), we would have up to about 10 drivers.  On slower weekday nights, we would have 3-4.  Drivers would also not usually work 8 hour stretches.  Drivers would typically be promised a minimum of 4 hours, and then those who wanted to leave earlier could (if business was slow).  So the franchisee could adjust labor pretty quickly.  With self driving vehicles, you've got a constant expense, can't let a robot go.

[DTEJD1997]

I'm in the US, Colorado right now. Your right that it may have just been consumer research disguised as a legitimate service, it crossed my mind. Another angle I was thinking about is how much of an incentive is it for the customer (if given the choice) to pick automation where no tip is involved or expected? As it stands now I have to pay a delivery charge and tip also. Maybe it will take longer to implement then I think. Just to examine it a bit more, quick google says the average domino's sells about 132 pizzas a day, I don't know what % is delivery, maybe 80%? So 105 a day, maybe allocate 25% of the volume to the automation to start, so 26, with a 5 dollar delivery charge, 130 a day? Over a week 920, a month 3680. Seems like it could pay for itself, maybe even increase profitability as you would depreciate it and could also deduct interest off a loan if it was financed? Obviously it would be a franchise by franchise decision, and your volumes would have to support it, but I bet for the right locations it could boost profitability.

 

UNF2007:

 

When I worked at Domino's there were people that picked up the pizza in person.  I would put that at less than 5%,  Domino's selling point was free delivery in 30 minutes or less.

 

Back in the 90's, there was no delivery charge per se.  The cost of delivery was built into the price of the pizza.  I just called Domino's and it is $3.50 charge for 2 larges in my area, it is waived "if you order a bunch of pizzas".

 

So maybe I stand corrected, and there might indeed be a market for the self driving cars.

[Jurgis]

https://www.nytimes.com/2018/11/18/technology/artificial-intelligence-language.html

[Jurgis]

https://www.wired.com/story/googles-ai-guru-computers-think-more-like-brains/

 

Short and somewhat info-lite article. But has couple interesting thoughts that are important (and I agree with  ;))

 

GH: I’m an expert on trying to get the technology to work, not an expert on social policy. One place where I do have technical expertise that’s relevant is [whether] regulators should insist that you can explain how your AI system works. I think that would be a complete disaster.

 

People can’t explain how they work, for most of the things they do. When you hire somebody, the decision is based on all sorts of things you can quantify, and then all sorts of gut feelings. People have no idea how they do that. If you ask them to explain their decision, you are forcing them to make up a story.

 

Neural nets have a similar problem. When you train a neural net, it will learn a billion numbers that represent the knowledge it has extracted from the training data. If you put in an image, out comes the right decision, say, whether this was a pedestrian or not. But if you ask “Why did it think that?” well if there were any simple rules for deciding whether an image contains a pedestrian or not, it would have been a solved problem ages ago.

 

Two thumbs up. BTW, what he said about people making up a story is also applicable to investing ... and investment forum discussions.  ;)

[Jurgis]

https://www.nytimes.com/2018/12/26/science/chess-artificial-intelligence.html - A bit glowy but makes some good observations about AlphaZero vs. older rule based chess programs.

[Jurgis]

Here goes (part of) Starcraft II: https://www.pcmag.com/news/366177/ai-dominates-human-professional-players-in-starcraft-ii?amp=1

[Jurgis]

Pre-cancer detection: https://www.usnews.com/news/health-news/articles/2019-01-10/ai-beats-humans-at-detecting-cervical-precancers

[neil9327]

An interesting thread. One thing I notice is that the AI algorithms discussed don't appear to try to understand the behaviour of the markets in terms of the trading strategies of the market participants. Instead they try to predict future prices based purely on past price information. In my mind this is surely putting the cart before the horse, in the sense that in the markets, price and price changes are what result FROM the actions of market participants, not the other way round. Yes it is true that the actions of market participants are often driven by price, but it is always historical price information that is taken into consideration, even if it is a few milliseconds ago - the current market price is never precisely known until the order is placed and the trade confirmed.

 

The price at any moment in time is always the price at which buy and sell volume are exactly matched - if it is not, the price moves up and down in an instant (thanks to high-frequency traders) until it is exactly matched.

 

So if you could model the behaviour of market participants in terms of what volume they would each add (if they buy) or subtract (if they sell), then you could model the future of price changes.

 

Now of course this is very difficult because peoples' trading strategies are often complicated, chaotic, and subject to emotional influence, but it occurs to me that many new traders in particular are likely to be using simple trading strategies based on popular technical analysis methods, and similar. If we wanted to model this behaviour using AI we could potentially do this. And if this model was able to do this successfully then we would be in a better position to use AI to go on to model the prices that are more likely to occur as a result of this behaviour.

[cameronfen]

^^ Yes but you can make a lot of money in the stock market with good predictions without understanding why those predictions are made.  I doubt many professional pool players understand the underlying physics of the system, yet they can tell you what happens. 

[Jurgis]

https://techcrunch.com/2018/12/31/this-clever-ai-hid-data-from-its-creators-to-cheat-at-its-appointed-task/ - Fun stuff

[Liberty]

https://a16z.com/2019/02/22/humanity-ai-better-together/

[Liberty]

https://pathak22.github.io/large-scale-curiosity/resources/largeScaleCuriosity2018.pdf

 

Fascinating machine learning paper on self-supervision and intrinsic motivation (aka "curiosity") for AI agents.

[SharperDingaan]

An interesting thread. One thing I notice is that the AI algorithms discussed don't appear to try to understand the behaviour of the markets in terms of the trading strategies of the market participants. Instead they try to predict future prices based purely on past price information. In my mind this is surely putting the cart before the horse, in the sense that in the markets, price and price changes are what result FROM the actions of market participants, not the other way round. Yes it is true that the actions of market participants are often driven by price, but it is always historical price information that is taken into consideration, even if it is a few milliseconds ago - the current market price is never precisely known until the order is placed and the trade confirmed.

 

The price at any moment in time is always the price at which buy and sell volume are exactly matched - if it is not, the price moves up and down in an instant (thanks to high-frequency traders) until it is exactly matched.

 

So if you could model the behaviour of market participants in terms of what volume they would each add (if they buy) or subtract (if they sell), then you could model the future of price changes.

 

Now of course this is very difficult because peoples' trading strategies are often complicated, chaotic, and subject to emotional influence, but it occurs to me that many new traders in particular are likely to be using simple trading strategies based on popular technical analysis methods, and similar. If we wanted to model this behaviour using AI we could potentially do this. And if this model was able to do this successfully then we would be in a better position to use AI to go on to model the prices that are more likely to occur as a result of this behaviour.

 

AI algorithms actually are just an application of technical analysis, but in a diiferent context.

'History can predict future events'; in tech speak, make the machine calculate all possible correlations in a data set - & it WILL find some that are 'somewhat' predictive (middling R-square values). As it applies these correlations, we call it 'learning'. Of course, the 'machine' is only as 'smart' as the R-square of the correlation, and it's stability in an out-of-sample application; introduce it to a market-discontinuity, and it goes beserk :)

 

One of the theoretical arguments around HFT is that if your holding period is very small (nano-seconds), almost all your price gain will be attributable to market drift; and we can calculate the amount of that drift, using the Brownian Motion equations. Applied to AI, the more you can apply the Brownian Motion equations to an AI algorithm, the more accurate and stable it becomes. 

 

All things coming out of the 'investment' silo, and making the jump into other places.

 

SD