The Entrepreneur Forum | Financial Freedom | Starting a Business | Motivation | Money | Success

Welcome to the only entrepreneur forum dedicated to building life-changing wealth.

Build a Fastlane business. Earn real financial freedom. Join free.

Join over 80,000 entrepreneurs who have rejected the paradigm of mediocrity and said "NO!" to underpaid jobs, ascetic frugality, and suffocating savings rituals— learn how to build a Fastlane business that pays both freedom and lifestyle affluence.

Free registration at the forum removes this block.

The AI Entrepreneurship thread. Code AI? Interested in AI? Data Science? Machine Learning?

ChrisV

Legendary Contributor
EPIC CONTRIBUTOR
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
225%
May 10, 2015
3,141
7,055
Islands of Calleja
I am into medical science. So I am actually planning to make learning for medical students easy and fun
If you come up with a course I'd love to check it out. Also check out this channel. This guy is literally the best teacher I've ever seen in any subject, ever. I spoke to him and convinced him to make a few videos outlining his teaching method. The way this guy teaches is revolutionary.

 

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
Anyone knows a good book to read about ML? Recently I read The Hundred-Page Machine Learning Book, and I am starting Machine Learning Yearning by Andrew Ng.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.
D

Deleted74925

Guest
Anyone knows a good book to read about ML? Recently I read The Hundred-Page Machine Learning Book, and I am starting Machine Learning Yearning by Andrew Ng.

A good resource that's based in R: Hands-On Machine Learning with R

For Python users, this is probably best: Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems: Géron, Aurélien: 9781492032649: Amazon.com: Books

Honestly, I wish that I had started learning in Python
 

ChrisV

Legendary Contributor
EPIC CONTRIBUTOR
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
225%
May 10, 2015
3,141
7,055
Islands of Calleja
Honestly, I wish that I had started learning in Python
Yea, I'm an R coder and I really wish I had learned python as well.

Don't get me wrong, I love R but the way the industry is going you're probably going to be in a much better position if you know python.

These books are also great too:



You can also find their lectures on EdX / YouTube.
 

1_4m_P4tty

New Contributor
User Power
Value/Post Ratio
25%
Apr 2, 2020
4
1
Philippines
If you come up with a course I'd love to check it out. Also check out this channel. This guy is literally the best teacher I've ever seen in any subject, ever. I spoke to him and convinced him to make a few videos outlining his teaching method. The way this guy teaches is revolutionary.

Thanks for sharing :) I'd like to come up with a program that will help consolidate important information that is needed to ensure quality patient care. My observation is that many medical students graduate still clueless what to do in their field. I am all for learning all the basic and pertinent stuff so that when you go out there and do your job, you are well equipped with a solid knowledge base that could help you predict possible problems, how to avoid them, how to fix them and how to improve them.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina

MHP368

Gold Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
161%
Aug 17, 2016
794
1,277
37
Sahuarita AZ
"Data Driven: Harnessing Data and AI to Reinvent Customer Engagement"

good primer on this (from a marketing perspective) , its what i'm doing, more as a "grab the coat tails" play then any specific genius maverick move on my part but I firmly believe AI / IoT is going to make more millionaires and billionaires in the coming decade then anything else on the scene.

besides the company I resell for : 6sense, Bombora, Intent Data IO, and LiveRamp are the only real players in this space.

The play is "intent" or behavioral marketing. So , when you google "emergency plumber dublin" you're letting google know you have probably a rather immediate need for an emergency plumber in dublin right?

Now lets frame out the path to purchase on something big time, a house or car. Well you probably do research right? so it wouldn't make sense to slam you with ads for a black jeep right when you still haven't decided between a car or a truck or a motorcycle now would it?

but those signals are still obvious, what happens when we examine the url level behavior (and any other data signals allowed with IoT), thousands of datapoints a day (a japanese researcher figured out a way to make a personalized fingerpirnt from the way you sit in a chair, hows that for unique personal data?) on hundreds of people who have compelted the path to purchase? what correlates do we see?

So you feed all this data into a machine elarning algo and tell it to look for patterns and voila, we see smoke , we see fire. Now i'm not going to sit here and pretend we dont need to eke out causation just because we have a lot of correlates but uh, i'll let some grad student pour over that junk later, right now I'll just make money on it.

What about an enterprise level transaction/ what if I could tell you that the Vp of marketing at the company you want as a client is considering purchase of your product before he's even consciousely aware of the fact? , what if we figure out a signal that would allow us to know in advance before a company moving millions around was even starting to consider a purchase? before it even occurs to accounting or gets to the CEO that a solution is even needed or a problem exists?

Why you could, open up old communication channels, forge new ones, maybe send a "just because" bottle of wine. If you're the first person in front of a customer you get frame control, now every other possible seller is competing in the frame you set , whats that worth?


Really exciting stuff, I mean what i'm talking about is just one application but lots of money is pouring into getting the entire product experience synchronized, no more ad spend on someone whos already purchased. Just enough ads on just enough channels (omnichannel marketing) to push you over the edge to purchase, tailoring individual ads (what if your smart tv's commercials started using your first name? , thats an easy one in the works even right now) , as more devices become internet connected and send more signals the possibilities just boggle. Whats a beverage company going to look like once they have biometric data about what the first taste does to you? , lots of crazy interesting stuff in the pipeline.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

MHP368

Gold Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
161%
Aug 17, 2016
794
1,277
37
Sahuarita AZ
I have a feeling that we're going to be talking about a bubble in the AI space in a couple years...

Agreed, some of the stuff is just sloppy application of API integration with "AI" slapped on to get funding.
 

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
The other day I saw an interesting video about ML. It's an interview about one guy who sold his company.

The key point, I guess, is that we cannot expect to create a SaaS only about ML, but a some features with the technology (small maybe) and the rest of the system like any other.

In his case, he used a simple clustering, nothing fancy.

View: https://www.youtube.com/watch?v=fB7nyxXaczY&list=WL&index=3&t=0s
 

Julio Andres

Contributor
Read Fastlane!
Read Unscripted!
User Power
Value/Post Ratio
128%
Dec 9, 2014
40
51
38
Following! I'm a software engineer, mainly app developer, but interested on entering AI-ML field because it sounds fun. I'm also taking the coursera deep learning courses, and looking into kaggle.com, it seems a pretty well know site for data science-ML practitioners.

And always thinking on how to apply this to some business. I'm searching on kaggle.com for interesting datasets, and get inspiration for a business.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.
D

Deleted74925

Guest
Also check out this channel. This guy is literally the best teacher I've ever seen in any subject, ever. I spoke to him and convinced him to make a few videos outlining his teaching method. The way this guy teaches is revolutionary.


I link to StatQuest in my free ebook on machine learning. I messaged the author and asked for permission, which he was ok with.

One of the people taking my course likes to debate. Yesterday he began trying to convince me that Root Mean Squared Error is a bad metric to use and that AIC/BIC is better. I had to dig out my old textbook (actually just google it) to answer the question.

Do I think that knowing all of this math is necessary? Not for 99% of people. The only reason that I learned this was because it was required for my job. It's useful for doing academic research, if teaching (in my case), or if working in a ML engineering role (in my old job). But many people today are just amassing the knowledge for no reason. They take online courses on Coursera/Udemy/Kaggle endlessly, for no purpose, only so that they can add it to their resume.

This is from our Moodle discussion forum:

33502

33503

33504
 

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
Maybe someone can help me with this issue.

I have been playing with a DQN (rainbow) and NES games in and old computer (I5 4670, 8 GB and a 960 GTX 2 GB) because it's the only with a graphic card. I don't expect so much, but at least is faster than my Apple computer for ML task.

Lately I've been thinking to make an upgrade and I try to find what to upgrade first: CPU, GPU, RAM. So I saw my CPU is 100% while my GPU is around 50-60%. I made some research and I found that the DQN use mainly CPU, because the only part of the code that use GPU is the NN, and because is small it's not going to benefits for a better GPU.

While the replay memory is filling the CPU is around 50%, but when it starts to send batch to the NN it jumps to 100%. Also, the FPS down from 180 to 30

So if I buy a CPU (I've thinking a 3800X with 16 GB) I am going to see some improvement, but in the future when I upgrade the GPU is going to be the same. I mean, I can spend money but at the end the improvement is not going to be so great.

I am missing something?. At first, I thought that maybe the problem was lack of RAM because SO has to paginate, but with a smaller replay memory the same thing happens. Another thing that I thought it could be is the VRAM, but at least TensorFlow should not use RAM (The code is in PyTorch, but I think is kind of the same).

I wonder if with a better PC part I will see a significant improvement (various magnitudes) because there are something (like RAM or VRAM) that made the CPU work harder than it should. Or the CPU is going to be always the really bottleneck and if I don't but a Threadripper I am not going to see a real improvement.
 

Christygruber124

Contributor
Read Fastlane!
Read Unscripted!
User Power
Value/Post Ratio
167%
Jul 16, 2020
12
20
PA
What’s the difference between machine learning, AI, and data science?

what would be a good intro course to get an understanding of the two to see if it’s something I want to Persue? Should I learn coding before I look into these subjects?

You are right there are numerous opportunities for machine learning AI and so forth. What it could do for people with disabilities is phenomenal. An automated car so that the blind don’t have to depend on others to get them places. A device that translate the spoken word Perfectly to text for the deaf. It’s endless!

Its funny how people fear that it will make jobs obsolete. No it just means new opportunities if we keep our eyes opened.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

Owner2Millions

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
103%
Feb 21, 2017
144
148
Chicago
@ChrisV Are you going do another AI project this year? Im just now seeing this thread. subscribing to be notified on this thread.

I realize there is a lot of demand for AI and ML, especially for Robotics and Drones. So system engineers are going be in demand for a while and highly paid. But it has always been lucrative to learn the hardware and software side of tech. IMO

Everyone is launching SAAS apps for particular fields. It seems to be a bit oversaturated. I know this isn't the peak yet. But, I am seeing a ton of companies within SAAS looking like the bread aisle at Walmart lol.
 

ZF Lee

Legendary Contributor
EPIC CONTRIBUTOR
Read Rat-Race Escape!
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
180%
Jul 27, 2016
2,840
5,113
25
Malaysia
What’s the difference between machine learning, AI, and data science?

what would be a good intro course to get an understanding of the two to see if it’s something I want to Persue? Should I learn coding before I look into these subjects?

You are right there are numerous opportunities for machine learning AI and so forth. What it could do for people with disabilities is phenomenal. An automated car so that the blind don’t have to depend on others to get them places. A device that translate the spoken word Perfectly to text for the deaf. It’s endless!

Its funny how people fear that it will make jobs obsolete. No it just means new opportunities if we keep our eyes opened.
Data science basically involves discovering, interpreting and forecasting new info from masses of data, be it organised or not.

When I started out in analytics in university, I covered areas like:
  • Regression modelling
  • Forecasting
  • Data Mining
And so on.

Machine learning is simply training a mathematical model to produce future predictions without needing explicit programing. AI merely applies some of ML, as there's other stuff it needs to simulate human behaviour.

DataCamp has lots of great intro and advanced courses to start with.
But you should understand basic statistical principles like:

  • Hypothesis testing
  • Type 1 & Type 2 errors
  • Common biases, estimators and errors (e.g. RMSE)
  • The steps of data modelling
  • Sampling & cross-validation methods-and how they work
Data Camp already has basic courses for them.

After all, statistics is the very root of data science and ML.

I learned up data science with R, as it is more intuitive than say, Python, and somewhat less code-heavy.
So you can try learning it first.
But mind that R's UI can be quite outdated.
 

lowtek

Legendary Contributor
FASTLANE INSIDER
EPIC CONTRIBUTOR
Read Fastlane!
Summit Attendee
Speedway Pass
User Power
Value/Post Ratio
332%
Oct 3, 2015
2,164
7,186
42
Phoenix, AZ
Maybe someone can help me with this issue.

I have been playing with a DQN (rainbow) and NES games in and old computer (I5 4670, 8 GB and a 960 GTX 2 GB) because it's the only with a graphic card. I don't expect so much, but at least is faster than my Apple computer for ML task.

Lately I've been thinking to make an upgrade and I try to find what to upgrade first: CPU, GPU, RAM. So I saw my CPU is 100% while my GPU is around 50-60%. I made some research and I found that the DQN use mainly CPU, because the only part of the code that use GPU is the NN, and because is small it's not going to benefits for a better GPU.

While the replay memory is filling the CPU is around 50%, but when it starts to send batch to the NN it jumps to 100%. Also, the FPS down from 180 to 30

So if I buy a CPU (I've thinking a 3800X with 16 GB) I am going to see some improvement, but in the future when I upgrade the GPU is going to be the same. I mean, I can spend money but at the end the improvement is not going to be so great.

I am missing something?. At first, I thought that maybe the problem was lack of RAM because SO has to paginate, but with a smaller replay memory the same thing happens. Another thing that I thought it could be is the VRAM, but at least TensorFlow should not use RAM (The code is in PyTorch, but I think is kind of the same).

I wonder if with a better PC part I will see a significant improvement (various magnitudes) because there are something (like RAM or VRAM) that made the CPU work harder than it should. Or the CPU is going to be always the really bottleneck and if I don't but a Threadripper I am not going to see a real improvement.

I'm not an expert, but I do run a fairly beefy system. 2x RTX 2080Ti, i7 7820x, 96gb ram. I run a YouTube channel and teach some Udemy courses on reinforcement learning. Here's what I've found.

You have a couple things going on: the environment simulations are running on the CPU and then shuttling the data to the GPU for the calculations. The slowest piece of hardware is generally going to be the limiting factor, and I'll tell you how to figure out which in a minute.

Memory:

The replay memory lives in your RAM, while the neural network is going to live on your vram.

If you're doing convolutions, it's going to scale with the number of filters (and the screen image sizes you're trying to process), so vram usage goes up quickly as the number of CNN layers / size of screen images / number of filters increases.

You can mitigate the RAM usage, to some extent, by using numpy arrays and carefully choosing the data types you use for your replay memory. You don't need int64 to store actions, for instance. You probably don't need fp64 to store your screen images, either (assuming you're normalizing between 0 and 1). Your terminal flags can be stored as boolean, and this makes setting the Q for the terminal states trivial:
Q_value_for_state_s_t+1[dones] = 0

Processor:
I haven't implemented Rainbow, but if it is like regular DQN In the sense that it's running a single environment at a time, then increasing the # of threads isn't going to help. Your performance will be dominated by the single threaded performance of your processor. Intel is generally ahead in that regard, though I would argue that the better multithreaded performance of AMD (in nearly any other task) is going to be an improvement in general system usage. I wouldn't go with threadripper. I would go with the fastest Ryzen I could afford, as 12-16 threads is going to usually be enough and you'll have some thermal headroom such that your system won't heat up your entire room (my office is easily 5 degrees hotter than anywhere else in the house).

GPU:
You're correct that the neural network sizes are pretty small and don't generally max out your GPU. This changes if you're using convolutions on large batches of buffered screen images, but for a simple neural net it's going to be pretty easy on the hardware.

I didn't see significant changes in run times (maybe a 15-20% improvement) going from a 1080Ti to a 2080Ti, which is a better metric than the utilization of your gpu. I saw a much bigger improvement going from a GTX780 to my 1080Ti. In hindsight, I wish I had bought 2 due to the run up in prices after the crypto mania.

The biggest benefit you can get is to run 2 reasonably fast gpus, with each running a different set of agent hyperparameters. This lets you double the speed of the search of the parameter space, and will ultimately save you more time than a single top tier gpu. You can even run different algorithms (i.e. rainbow and D3QN) to compare performance.

Power supply: not something you mentioned, but this is the single most critical component of the system. If you don't supply adequate current on the 12V rails you risk frying your hardware. Buy Corsair. They make high end power supplies at a reasonable price point. I would also advise paying extra for the fully modular ones, as they make cable management a breeze (important for thermal management).

Cooling: Also quite critical. Modern cpus and gpus are set up to run at the max frequency their thermal management will allow. Better cooling means better performance. Pay the extra money for the all in 1 liquid CPU coolers and set up a regular schedule for cleaning out the dust bunnies. Make sure to set the CPU cooler fans up to exhaust the hot air out of the case.

So how do you know what the limiting factor in your rig is?

You're setting up your code something like this:
Edit: sorry for the poor formatting. The forum software removes the 4 space indent.

while episode is not done:
choose_action(state)
new state, reward, done, info = env.step(action)
store_transition_in_memory(state, new_state, reward, action, done)
state = new_state
call_learning_function_for_agent()
print_episode_debug_info_to_terminal

You can figure out where the bottleneck is by adding in:

import time
inference_time = 0
train_time = 0
env_time = 0

while episode is not done:
inference_start = time.time()
choose_action(state)
inference_time += time.time() - inference_start
env_start = time.time()
new state, reward, done, info = env.step(action)
env_time += time.time() - env_start
store_transition_in_memory(state, new_state, reward, action, done)
state = new_state
train_start = time.time()
call_learning_function_for_agent()
train_time += time.time() - train_start
print_episode_debug_info_to_terminal
print_times_to_terminal_after_1000_episodes

This will give you some clue as to where the bottleneck is. If the sum of the inference and train time is greater than the env time, then your gpus are taking up most of the execution time and an upgrade could benefit you (up to a point).

Hope that helps!
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
This will give you some clue as to where the bottleneck is. If the sum of the inference and train time is greater than the env time, then your gpus are taking up most of the execution time and an upgrade could benefit you (up to a point).

Hope that helps!

That's a really good answer, thanks Phil. I know your channel and I followed some of your tutorial in the past, also your video about how to spec a deep learning PC.

I run the code, and the inference and train time is greater that the env time (specially the train time), so the logic dictate that I should upgrade the GPU. But the task manager shows the CPU is at 100% all the time. So I don't get it, if I buy a better GPU the CPU is not going to "keep the pace" and I am not going to see any improvement. I am missing something here?

By the way, you set-up is amazing.
 

lowtek

Legendary Contributor
FASTLANE INSIDER
EPIC CONTRIBUTOR
Read Fastlane!
Summit Attendee
Speedway Pass
User Power
Value/Post Ratio
332%
Oct 3, 2015
2,164
7,186
42
Phoenix, AZ
That's a really good answer, thanks Phil. I know your channel and I followed some of your tutorial in the past, also your video about how to spec a deep learning PC.

I run the code, and the inference and train time is greater that the env time (specially the train time), so the logic dictate that I should upgrade the GPU. But the task manager shows the CPU is at 100% all the time. So I don't get it, if I buy a better GPU the CPU is not going to "keep the pace" and I am not going to see any improvement. I am missing something here?

By the way, you set-up is amazing.

HMMMM.

Yeah, the cpu could very well bottleneck a faster GPU. You see this quite often in gaming. It looks like you're running a haswell CPU, so just about any modern cpu is going to be faster.

Do you know how to overclock? If you have the thermal headroom you could crank up the cpu frequency and see if it helps. If it does, there's your answer.
 

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
HMMMM.

Yeah, the cpu could very well bottleneck a faster GPU. You see this quite often in gaming. It looks like you're running a haswell CPU, so just about any modern cpu is going to be faster.

Do you know how to overclock? If you have the thermal headroom you could crank up the cpu frequency and see if it helps. If it does, there's your answer.

Is the normal (not K version) so it is locked to overclock. But what I try was to run the code in an Apple Mac Mini with a newer I5 with 6 cores, and it was slower. Of course in the training part was because the lack of a discrete GPU, but also at the beginning while it was filling the replay memory.

So I guest is the GPU my first problem right now. At the end I am going to upgrade everything, because I could use a bigger buffer too. And upgrade a DDR3 it not worth the money, there are much expensive than DDR4.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

lowtek

Legendary Contributor
FASTLANE INSIDER
EPIC CONTRIBUTOR
Read Fastlane!
Summit Attendee
Speedway Pass
User Power
Value/Post Ratio
332%
Oct 3, 2015
2,164
7,186
42
Phoenix, AZ
Is the normal (not K version) so it is locked to overclock. But what I try was to run the code in an Apple Mac Mini with a newer I5 with 6 cores, and it was slower. Of course in the training part was because the lack of a discrete GPU, but also at the beginning while it was filling the replay memory.

So I guest is the GPU my first problem right now. At the end I am going to upgrade everything, because I could use a bigger buffer too. And upgrade a DDR3 it not worth the money, there are much expensive than DDR4.

The laptop cpu is probably running at a significantly lower frequency. It's pretty hard to compare across generations of CPU, as you get architectural enhancements. Those apple laptops are also pretty bad for thermal management, so you were probably throttling the cpu. Who knows.

I'd say a 7 year old cpu warrants an upgrade. I would look for used parts on ebay, if cash is tight.
 

Martin.G

Bronze Contributor
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
107%
Jan 31, 2019
175
187
Argentina
The laptop cpu is probably running at a significantly lower frequency. It's pretty hard to compare across generations of CPU, as you get architectural enhancements. Those apple laptops are also pretty bad for thermal management, so you were probably throttling the cpu. Who knows.

I'd say a 7 year old cpu warrants an upgrade. I would look for used parts on ebay, if cash is tight.

It's a Mac mini, so it's not the laptop version. I am going to try to install Ubuntu because the code is running Windows. I read that it can improve performance https://github.com/pytorch/pytorch/issues/22083

My goal is to upgrade everything, but in Argentina is kind expensive and difficult to get pc parts. For example, if I decide to upgrade a GPU I don't know if I should wait the new 3000 serie in September or buy a 2000 now, because prices and stock can change drastically in a week or month.

Edit:
With Ubuntu I have an improvement of almost 2x and the CPU utilization is 150% of 400%. Also the GPU jump to 80-90% utilization.
 
Last edited:

Kimblis

New Contributor
User Power
Value/Post Ratio
38%
Aug 4, 2020
8
3
Lithuania
So guys where do you see industry going in the upcoming years? Is it worth to dig deep into AI/Data Science if you will want to create something (considering you aren't a huge company like Google, Amazon etc.) and you're building from a ground up. Or is it better to go more into creating hardware/IOT stuff that will be needed for computation and also to gather data? If you would decide now to study one of them, which one you would go to? :)
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.
D

Deleted74925

Guest
So guys where do you see industry going in the upcoming years? Is it worth to dig deep into AI/Data Science if you will want to create something (considering you aren't a huge company like Google, Amazon etc.) and you're building from a ground up. Or is it better to go more into creating hardware/IOT stuff that will be needed for computation and also to gather data? If you would decide now to study one of them, which one you would go to? :)

The "frontiers" of AI are light years ahead of what you and I can access from online courses. Source: I manage an online course and it's taken me months just to create tutorials. I've heard of a few successful AI startups but they mostly get VC funding and have founders who went to Ivy league schools.

A more "fastlane" approach is to feed the crowd. What are the needs that all of these AI companies and people will need? All of these are unattractive, which means that there's less competition.

Skills

Sales and marketing of machine learning models
  • How can you prove that your model is worth $100/customer? Data scientists tend to come from technical backgrounds. I worked as a predictive modeler who would create the models and PowerPoint decks, but then we'd turn them over to our sales team who would sell them to clients.
  • My current course has about 30% of the curriculum just on communication. Basic things like writing, combining graphs/statistics into understandable language. The person who can explain their algorithm/AI wizardry in simpler language is worth more money than the person who struggles with communication.
Risks to fraud, data breaches, technical errors
  • Data cleaning, integrity, and reducing error. How can you know that your data is going to be consistent? "Garbage In; Garbage out." Data scientists need to know this, but it's not new! It's been around in Database language since SQL was invented in the 1980s... but people don't want to learn it because it's "boring".
  • Writing documentation.... Ugh. Is there anything that a data scientist hates more than doing this? Probably not. This task tends to get outsources to junior team members because it sucks. But the probme is that only the "expert" who created the model is knowleagable enough to write good documentation.
Hardware
  • How to use cloud services. There are basically three big competitors - Microsoft Azure, Google Cloud, and Amazon Web Services.
    • Companies concerned about data privacy still are hesitent to use the cloud, unless they are massive and have boat loads of money to spend on R&D.
    • Getting models into "Production". It's easy for someone to take their Python or R and make some AI tool, but it takes a lot longer to get it ready to be used in real-time. Most of the time this needs to be translated from R/Python into some other framework like JavaScript/Scale (?), Julia, etc. This is one of those tasks that data scientists hate.
    • Budgeting... The cloud offers "unlimited scalability" but with that comes unlimited costs. At my former company, we ran into major problems with cost overruns. We were using a machine learning model which was designed to run on about 100 GB of data, but then we had a client that had 100TB of data. The cloud computing costs went from about 20% of the costs (for normal clients) to 60% of the costs. This could have been avoided if we had knowledge of how to manage costs.
Hiring/Staffing/Recruiting
  • How do we hire the person with the right credentials?
    • Hire only Phd's
    • Hire Master's degree people and hope that their degree is actually useful
    • Put them through a rigourous hiring process involving technical projects, background checks with previous managers, reviews of their github/kaggle projects, and numerous on-site interviews and "white board" questions
  • How do we know if a "data scientist" from India who is selling themselves on Upwork for $500/hr is legit or not?
Education
  • Analytics managers have "continuing education" requirements like doctor's do. As algorithms get more complicated, it becomes more difficult to keep staff up to date. There's a growing need for corporate training. Remember all of that "Data Privacy" training that you were required to take? Imagine "AI Privacy" training in 3 years.
  • Learning platforms - My site is build using Moodle, which is good because it has a ton of open source documentation.. but there are newer platforms which have more features such as zoom integration, templated content (think Wordpress) for quickly making new courses, and so forth.
 

ygtrhos

Bronze Contributor
FASTLANE INSIDER
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
135%
Dec 27, 2016
240
325
34
I did not read the whole thread and I have a question.

I am a 31 year old simulation engineer. Basically I do technical simulations.

For example, if you have a connector for a pump and want to calculate the loads on the bolts at the connector, I can calculate it.

Good thing is, that I do not actually need to click around. I can sit down and program something, which actually builds that model and creates the code. It has been a while, but I can program in Pyhton, MATLAB and C.

Now recently, I have seen a video that shows the capabilities of this new GPT-3 code. It has read all Wikipedia and books that have been published since Earth existed or whatever and now it can compose its own poems.

Heck, it also writes the code if you say "write me a code that rolls a dice with 20 sides and records the first 10 throws". Or "make a mockup of a website that is a photo gallery where you can like or comment to each photo" (Instagram) and it makes a mockup of Instagram in minutes.

This brought me to the idea that I can apply this in my own field. I cost 100k€ to my company and there are probably like 100k-300k guys like me.

If I really can integrate AI to my own field, I would be spectacularly rich. You would not need then hours of work and 20 engineers in an office, but 2-3 very qualified ones would be enough.

Practically this would be like "calculate connector loads with 6 X bolts under Y kN of bolt pretension and the connector's drawing is in Z file."

This is an objective that requires creativity, knowledge and experience, but it is not dependent on emotions or emphaty, so I think it is possible.

But I have zero experience with machine learning or AI. So I have some questions:

1. Where to start with AI where you can automatize such objectives? What should I learn or how should I qualify myself?

2. I see a big challenge: Whereas you can find a lot of HTML code or poems on the internet, there is limited work on coding such specialized engineering software because most of such codes rot away in archives of geeks like me. The community is very small and very much disconnected, so such codes to sample from are not easy to acquire, I believe.

I heard that the bigger the data size is, the better. Is the "artificial intelligence" and the data size linearly proportionate? Or exponentially?

I hope I can convey where I am going with those thoughts. My concern is that it might not be easy to feed the code enough data.

I would appreciate any answer. :)
 
Last edited:

sonny_1080

Creating a tool I want to use.
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
164%
Oct 30, 2019
496
815
Los Angeles
Attorney General Anne Milgram uses "Moneyball"-esque techniques to reform the criminal justice system, with great success.

Anne Milgram: Why smart statistics are the key to fighting crime


When she became the attorney general of New Jersey in 2007, Anne Milgram quickly discovered a few startling facts: not only did her team not really know who they were putting in jail, but they had no way of understanding if their decisions were actually making the public safer. And so began her ongoing, inspirational quest to bring data analytics and statistical analysis to the US criminal justice system.
That's the most inspiring thing I've heard today.
 
Dislike ads? Remove them and support the forum: Subscribe to Fastlane Insiders.

ChrisV

Legendary Contributor
EPIC CONTRIBUTOR
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
225%
May 10, 2015
3,141
7,055
Islands of Calleja
Great interactive Demo of how a Convolutional Neural Network works

 

ChrisV

Legendary Contributor
EPIC CONTRIBUTOR
Read Fastlane!
Read Unscripted!
Speedway Pass
User Power
Value/Post Ratio
225%
May 10, 2015
3,141
7,055
Islands of Calleja
Great interactive Demo of how a Convolutional Neural Network works

Two more interactive Demos on Decision Trees:

 

Post New Topic

Please SEARCH before posting.
Please select the BEST category.

Post new topic

Guest post submissions offered HERE.

Latest Posts

New Topics

Fastlane Insiders

View the forum AD FREE.
Private, unindexed content
Detailed process/execution threads
Ideas needing execution, more!

Join Fastlane Insiders.

Top