What are you working on?
For the record, CS != Python programming. Try Operating Systems, Compiler Design, Category Theory, Computer Security, Computer Vision, Robotics, Natural Language Processing, Machine Learning, Game Theory, Functional Programming, Computability Theory, Databases, etc.
>>7734467
Undergrad focusing on machine learning/theoretical shit and discrete math here. I honestly hated cs and wanted to go full discrete math before I started taking courses on machine learning
Where does everyone download their papers from for free?
Good news everyone
Deep Learning
An MIT Press book in preparation
https://goodfeli.github.io/dlbook/
I'm working on neural architectures that train without back propagation. RBM kind of stuff.
>>7734548
functional programming is a field of research in computer science. it doesn't mean just haskell elitism
>>7734474
what do u like about ML?
>>7734549
>you mean literally every neural network outside of textbooks
>>7734584
The applications are entertaining (natural language processing in particular) and I found everything about it really intuitive
>>7734600
are you gonna go for grad school then?
>>7734589
Sure...? I'm well aware of the existing research in this area, which is why I mentioned RBMs....
>>7734606
I might, unless I can leverage my ML related coursework into getting a position somewhere that would let me work with it. Not interested in developing enterprise software or apps desu.
>>7734609
>Not interested in developing enterprise software or apps desu.
same boat as you family. applied for cambridge's mphil this year and if i get accepted i might consider a phd there too
>>7734609
>tfw did multiple ML research projects as an undergrad
>tfw particpated in multiple poster competitions
>tfw developing enterprise software now
where did i go wrong...
>>7734619
You wanted money nig
>>7734619
you let your dreams be dreams, anon
>>7734609
Does anyone employ undegrads for ML positions?
>>7734477
Many good papers will be free on the institution's site or through the conference. Just Google the name.
I'm developing a language.
>dependent types
>linear types
>pure
>total
>fast
>compiles to LLVM
>no garbage collector
>FFI
http://cloudcv.org/vqa/
(not my project)
>>7734467
Building a rudimentary renderer akin to Marmoset Toolbag, Keyshot, Cycles, etc. Obviously I won't make something as polished and functional, but it's a fun project.
I'm a first year in CS, but have self-taught for a year or so. Programming (in C) that is, not really much CS. I basically just know "up to" as "complicated" as a binary search tree, merge sort, what have you. Nothing hard at all.
What should I read/apply myself to in order to get better? Reading about really complicated data structures and algos seems too much for me right now, but I'm having trouble finding that transition area. Any good books that aren't graduate level? Thanks.
I'm stupid, so can anyone explain me the correct answer(s) please? I'm thinking a) (cause of smaller numbers) and c) (because that's what wiki told me, but not sure actually why)
My CC that I'm using to transfer to UBC still uses matlab.
Should I just learn the bare minimum to do the assignments and assume I'll be using python later?
>>7734656
You should learn MATLAB properly because you'll run into it all the time.
Anyone active on Kaggle? Thoughts? Tips?
I'm working on some CNN.
The training is really slow but the results are satisfying.
>>7734701
What am I looking at?
>>7734704
Hidden layer of a convolutional neural network.
They are used for feature extraction in image classification.
>>7734653
It's c.
a) Is wrong because for a computer, computing a very small number (1e-10) and computing a very large number (1e+10) is not that different.
Scaling the features speed up the GD process. You can imagine walking down from a hill. In case you normalized the features, the hill will be more rounded. If the features are not normalized, one side of the hill will be stretched accordingly to the proportion between the greater feature and the small feature. Since the hill is stretched, you need to walk many more steps to reach the bottom.
>>7734640
Anyone knows of any resource that shows how others analyse data? A video or a some reading material where others are given a set of data, they decide on and explain preprocessing methods, then explain how they model a solution an why, etc. Maybe a podcast or a stream? Book?
>>7734640
>boy
shots fired at naoto
>>7734717
Makes sense, thanks anon. Also your answer was correct btw
how do i become senior researcher at Google Inc.?
>>7734474
You should focus on math before focusing on ML
>>7734770
bullshit, you just need basic maths tbqh
Thinking of writing a somewhat trivial haskell or scheme interpreter in C for fun..I have experience with C and Scheme but want to learn proper babby compiler stuff..Is it worth learning haskell (i.e. for interesting projects in the future) if I'm already familiar with functional programming or should I just save some time and make it for scheme? Yes, I realize that languages in general are mostly trivial in this day and age and it's easy to learn new ones quickly, but I just want some insight anyway
I'm thinking about compiling some sort of a guide and put it into a pastebin.
Should I? I don't think there are many people on /sci/ who actually care about specialized topic such as ML. Just look at the organic chem thread for example, it's barely survived (or even dead now).
>>7734824
>Should I?
Do it anon. We'll make this a thing with time
>>7734824
yes! do you have any idea how long I've been trying to get into ML without a practical guide? do it!
>>7734824
go for it
>>7734888
Just do Andrew Ng's course?
what does everyone think about http://www.thetalkingmachines.com/ ?
Also any podcasts about AI that you'd recommend?
>>7734964
technical podcasts btw i swear to god if i hear more "philosophers" talking the "ethics of AI" i'm gonna fucking punch someone in the face
>>7734964
>go to episode one
>WOMYN N MACHINE LEARNING WORKSHOP
>close tab
literally fuck off
>>7734960
Check out his credentials tho. Guy is a powerhouse
>>7734960
I heard you were talking shit.
>>7734467
Reading up on Cartesian Closed Categories.
>>7734467
>For the record, CS != Python programming.
This
Java programmers are CompScis too
>>7734642
http://4chan-science.wikia.com/wiki/Computer_Science_and_Engineering
>>7734467
So, how many times did you all fail calculus?
>>7735042
failed calc 2 once (technically dropped it, but was certainly going to fail), but that course was spivak calculus . All the rest got passed in one go.
>>7735042
>implying you know the pace of my numerical analysis class
this was first year, 4 weeks module, ending right before exam week
>>7734548
>I don't understand it so it must be shit
>>7735364
>implying that's impressive
You're embarrassing yourself
>>7735364
>Simultaneous linear equations
HOLY SHIT HOW DID YOU MANAGE?
>>7734467
Working on porting my 2D convolution processor design to 3D convolution for large kernels on an arty FPGA dev kit.
Learning scala too for subsymbolic learning
Pic related is the board that the 2D convolution processor was implemented on, it was part of a really amazing course. We designed the whole thing, including the PCB.
>>7735396
>your only class about calc
So you're doing partial derivatives (as applied and engineering-tier as this class is) without taking calc before? Yeah, nah. Let me take a look at your curriculum. This goes after calc1 and along / before calc2 I bet
>>7735392
I hope you're doing it FFT-based? Does it support bit-reversal at machine level?
>>7735384
(((((((((((((((((((((S)(O)((( )))(w)(o)(w)((((\n))))))(((S)(o)(( ))(t)(h)(e)(o)(r)(e)(t)(i)(c)(a)(l))))))))))))))))))))))))))))
>>7735457
>all functional languages are lisp
>le parentheses meme
>>7735438
>This goes after calc1 and along / before calc2 I bet
lmao yeah right. we were told to learn differentiation before class if we don't know what it is
>>7734632
>developing a language
Oh great, language #4090293.
>>7735483
if you are a CS student/grad and you've never developed your own language, please leave the thread. it's not about making a groundbreaking language, it's about helping you understand the issues of language design. any worthwhile degree has a course on this
>>7735479
Mathematical foundations doesn't include one-variable differentiation? What kind of shitty curriculum is this? Why are the class names so weird?
>>7735496
>>Mathematical foundations doesn't include one-variable differentiation?
nothing close to it
>Mathematical Foundations of Computer Science
>This module lays the foundations in discrete mathematics for many areas of computer science. It also explores techniques of mathematical proof, including so-called constructive proof where we show something exists by giving a method or algorithm to construct it.
Out of interest, how many of you ML enthusiasts are self taught?
If you are, what are you currently working on, how long did it take you to develop the depth of understanding necessary to reach where you are, and what skill set did you have prior to beginning learning?
>>7735500
>including so-called constructive proof where we show something exists by giving a method or algorithm to construct it.
Your professor sounds like an undergrad
>>7735528
>talking shit about one of the original designers of Haskell
bitch you don't know what's up
>>7734632
>pure
>dependent types
Kick ass!
>>7735524
I wouldn't call myself self-taught but I did self-teach myself Haskell when I chose to implement a compiler in that language instead of C++ like my classmates (I did this for a course, hence not really self-taught).
I'm actually a mathfag so I spend less time coding and more time doing theory. Currently learning about categorical logic, though I've begun to dip my toes into Agda. That said, I write stuff in Haskell whenever I get the chance, the language is just super intuitive once it clicks. I went into it with no background and basically learned by reading
https://www.umiacs.umd.edu/~hal/docs/daume02yaht.pdf
and then supplementing with
http://learnyouahaskell.com/
I think the major thing that helped was actually having a project I was learning it for. I probably would've abandoned it if I just sat down one day and decided I was going to learn this programming language for no reason.
>>7735528
>get a load of this pleb
Positive proof by contradiction (i.e. using double negation/law of the excluded middle) isn't really useful for implementation.
>>7735496
In Computer Science people are interested in computable functions. Formally these are functions from the integers to the integers (with a bunch of other non-trivial criteria, google 'general recursive functions'). The real numbers are an inaccessible fantasy world from the perspective of comp sci (at best you get floating point or fixed point arithmetic which doesn't behave nearly as nicely as the rational numbers).
Calculus is only of interest to comp sci people either when dealing with applications or when dealing with complexity.
>>7734795
Yes, definitely do it!
Most of compiler implementation deals with learning stuff about subsets of the set of context free grammars. Here is a book that covers the material with Haskell.
http://www.cse.chalmers.se/edu/year/2012/course/DAT150/lectures/plt-book.pdf
In my opinion it's missing some things and missing some explanation but it still isn't a bad book. It's worth mentioning that the book uses BNFC but one typically learns the theory for implementing a compiler manually before using that tool (it trivializes everything).
You can also use a C++ book for the Lex and Yacc stuff but adapt it to Alex and Happy (the Haskell equivalents). In my opinion it's a great learning experience.
>>7734824
Go for it.
![68b082518adb4ef4b32de5ea95e89ba7[1].png](https://i.imgur.com/8xksFsX.png "68b082518adb4ef4b32de5ea95e89ba7[1].png")
>>7734640
ebin
>>7735493
My uni has a non-mandatory course on making a language + compiler, I didn't even realize it wasn't part of the main programme cuz almost everyone seems to have taken it. Anyway looking forward to that shit.
if you're not doing AI you don't exist, bye.
>>7734640
hmm
>>7735834
no more training is needed, the net works great
I got its opinion on futanari.
These were the results.
Numbers don't lie, folks.
>>7734467
Biomedial signal processing using machine learning in Python.
>>7735931
Sounds cool, show us?
>>7734640
Was it right?
>>7735961
I still don't even know what the fuck was going on with this shit.
I'm pretty sure there was like a team of 1000 people who got together and decided that they were going to take to the internet and argue that this dress was a color that it clearly isn't, probably for a marketing ploy or something.
>>7735978
>being this dense
>>7735993
It's blue and black.
What the fuck.
Mfw I'm honestly too dumb or poor to concentrate into any useful field of CS. I graduated with a thesis on databases, and that eventually (after many years) gave me a job developing some php crap.
I was happy I was fired after failing to create some grand scale project as a one man project, with a rather unclear goal. Some minor projects were fun.
Now my days are filled with taking care of our toddler and very little time to self learn anything new. Must find some niche specialty some day.
>>7734467
>What are you working on?
Web dev for my part time job :^)
>every other poster claims to be an ML expert
>no one posting any projects
>>7736044
I saw this picture once before some weeks ago.
The illustration is campy, and gives a strange ret-conning kitsch to what is essentially a very simple speculation about the future in a children's book. We are cultured to appreciate such things ironically, especially given the un-serious context.
But upon actually reading the pages, it becomes clear that the basic predictions for how computers would come to affect society are largely correct, and so the image cannot be dismissed ironically as simple kitsch. The one optimistic prediction which is obviously wrong, is the one about "computer crime" getting harder and harder to perform.
Also the guy in the foreground has cp on that drive kek. Even the look of the cop's outfit isn't a far cry from FBI/SWAT personnel.
CS minor (EE major), haven't taken stats/probability but wanna get into machine learning. What books worked for you guys?
>>7736715
no books i just read and understand anything cause im more intelligent than everyone and books are harmful
>>7736873
I was hoping a non-autist would answer, but thanks anyways m80
>>7736879
mad asf that your low iq keeps u trapped in the deprecated age of books? nerd keep up the pace
>>7734640
>0.01% chance of a harem, nice.
Implying more than a handful of computer scientists can do higher category theory
>>7737137
>haskell community
>handful
>>7737140
>haskell community
>everyone can do higher category theory
yeah, no.
>>7737140
That's like learning what a power set is and saying you know set theory.
>>7737289
wait is there more to set theory?
>>7736044
>having children
rip your dreams of ever becoming elite
>>7737322
>what's a nanny
>what's a wife
Reminder that if you concentrate more on the technology than on the theory and maths you will be deprecated in 5 years max
>>7737330
Your wife is too busy cucking you for your failures
>>7737383
this isn't webdev kid
After 2 years of working with ML, I feel like math subfields such as topo, category theory would have some crucial elements that could help advance ML a lot faster. However, we will never know because no one who aced in those subfields would give a fuck about ML.
This is why ML has stuck in the neural network bandwagon. Not like it's a bad thing, but I think there should be better ideas than FUCKING MATRIX \MULTIPLICATIONS.
>>7737330
>what's leaving your kid in the car while you go shopping/to work/to the bar
honestly a much better solution.
>>7734467
>What are you working on?
Pic related, motorbike detection. It took the computer 15 seconds to process a single frame, I would say that's pretty slow. If a frame has a strong perspective distortion, it needs like 75 seconds to process properly. Even so, the recall rate isn't that great either (~80%).
I don't know anymore.
>>7737442
To be fair it would take a human a fair while to count all those motorbikes.
But we evidently have some kind of l33t hax that permit us to look at it and within a second surmise that there is a shitload of motorbikes in frame aligned in approximately the same direction.
Is there a way I can work with graphs (sets of edges and nodes) in python? Where should I start if I know no python? It's for an algorythmics class, and I want to avoid using edgy (shitty scratch clone).
>>7737463
>But we evidently have some kind of l33t hax
This is stupidly true. Machine is still lacking in the regard.
>>7737409
It's matrix multiplication because your code is vectorized. That means that it's faster than fucking for loops.
I'm into category theory and ML but I don't give a fuck about probability so you should add that to your list of pre-reqs. I did have a friend who did his dissertation on a bunch of topology on clustering algorithm shit. It sounded pretty cool but I never really looked into it beyond layman explanations he gave.
>>7737472
MIT OCW has lectures on graph theory using Python.
>>7737486
Thanks family.
Do you mean these vidoes (what came up first in google)?
http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-042j-mathematics-for-computer-science-fall-2010/video-lectures/lecture-6-graph-theory-and-coloring/
>>7734467
Hello I'm a software engineer and I'm just trying to pass the semester. However, seeing as it's my first semester, I still have lots to learn in order to work on something serious. Have a good day.
>>7737442
>Actually doing things
Go away, you are not welcomed here.
>>7735724
Thanks for that book; I've looked at the course related to it and whatnot and it seems like an acceptable start for learning about compilers, might use dragon book later for a more comprehensive approach after finishing this.
>>7737499
>Do you mean these vidoes (what came up first in google)?
Actually, more like these: http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-00sc-introduction-to-computer-science-and-programming-spring-2011/unit-3/lecture-21-using-graphs-to-model-problems-part-1/
I haven't watched your videos, so I don't know if they overlap. MIT's site is a bit of a mess, I'm afraid. I followed their course on EDX anyway, but the content is supposed to be the same as on OCW.
>>7737538
Thanks, that looks to be more like what I'm looking for.
>>7734467
>What are you working on?
Mandelbrot zooming renderer. I'm currently writing library for 128bit fixed point decimals with SSE optimizations because double-precision floats are not enough.
Also I'm writing emulator and assembler for fictional CPU architecture for assignment.
Pretty simple stuff but really fun.
>>7737476
Again, in fairness, our brain has been sculpted by many a year to identify at a glance people/animal shaped blobs, how many of them there are, and what they're doing.
Plus I'm actually fairly certain the brain edits sensory information on the fly to make it more easily analysed.
Evidence for claim:
You know how the interior of a house looks completely different the first time you visit it compared to how it looks when you've been there a few times?
>>7734467
programming python
>>7737554
Is it really not possible to write a fractal zoomer like that without resorting to very high precision number types? It seems kinda silly, can't you just reset the stuff in some way when you run out of precision?
>operating systems, compiler design
Those belong to computer engineering (and even EE in some schools) just as much as CS, if not more, these days.
>machine learning
Ah, that cousin of computational statistics that's missing its legs
>robotics
Pretty interdisciplinary. MechE's and EE's are playing roles that are just as important in this field
>game theory
That's straight up economics, m8
>natural language processing, computer vision
Pretty interdisciplinary, but I'll give those to CS
>category theory
While heavily associated with functional programming, it should belong to math more than CS
>functional programming, information security, computability theory, complexity theory
Those are CS in its truest form, the best kind of CS.
>>7737759
AFAIK it's not possible.
If you want to make more serious zooming like in many YouTube videos you will need lots of precision.
I suppose it's because when you calculate the color of Mandelbrot fractal you essentially square and add complex numbers, so the result can change in unpredictable manner of you ignore any bits from information about the number.
Also it's impossible or at least very hard to find repeats in Mandelbrot to fake deep zooming by rendering the same structures found in more shallow places.
>>7737442
thats really interesting. How are you developing this? what language and framework?
>>7737442
It missed one though.
>>7734619
I know this feel. I worked on robotics and computer vision in school. Graduated and now I develop enterprise software for an insurance company.
>>7737844
Looks like openCV. Probably using c++
Mostly interested in computability theory and computational complexity. Getting more into data structures and algorithms. Want to switch into something applied so I can secure industry jobs but not sure in what? Maybe databases. But theory is where my heart is
>>7734632
Write me an email,
I'd also love to do one, but just do have one.
>total
I don't know why that's necessary desu
>>7737832
CS is known to steal something from everywhere
>>7737442
Anon, can you run the image through a first-pass filter which marks all the things which are obviously not a bike to not be analysed?
You could probably shave off a few seconds if you just marked the road as 'not bike - don't search here'.
>>7738508
Most CS grads are an uninspired bunch. Developing enterprise software brings you stability, because of steady pay and lots of job availability, and isn't too challenging.
By contrast, the cool computer vision and machine learning jobs are a much harder to come by, compared to software dev jobs.
>>7738508
>>7739893
I went to a state school. I got scholarships to some high profile schools but because I wasn't quite good enough to get full rides, I couldn't afford to go. Had to take out loans just to go to a state school. When I was offered a job that paid 70k before I graduated, you bet your ass I took that job. How the fuck else would I pay my debt? A bit of experience isn't a bad thing. But I've been doing this enterprise software development for 1.5 years now and I grow tired of it. Going to start the job search up again soon.
>>7738490
That is almost certainly what he is already doing. He'a probably using a Haar cascade classifier thing. He's just got so many bikes that it still takes a while to determine if the non-road bits are actually bikes.
music to midi converter
im still doing concepts tho it's hard to understand music theory if you don't play an instrument
Sentiment analysis of tweets about presidential candidates.
>>7740130
how do you come up with sentiment categories?
>>7740143
supervised learning.
>>7735457
*tips hat*
>>7740122
What do you need to know about music theory?
>>7736004
You're still talking about it. You've walked right into their trap.
>>7734467
What is the next step after learning a programming language?
Is machine learning viable?
What are the prerequisites?
>>7740219
Boolean Algebra, Discrete Math, Linear Algebra, Statistics
>>7734548
>I'm used to programming silly state changes rather than expressing processes as mathematical expressions.
What kind of cool math can you learn along with Haskell? I'm asking as a math/programming noob that wants to do some cool stuff.
>>7735457
Nice fedora senpai
>>7740247
Lambda calculus.
>>7740089
Oh that's a bit disappointing.
I really enjoy hackerrank.com because it allows me to practice for my database systems class. It has everything important: relational algebra, basic sql, more stuff like sub queries and etc, even some nosql stuff although I didnt have that class at college yet.
You can also practice java etc I guess, but I'm in the process of studying databases and it's perfect
>>7737832
Programming language theory is a really big area of CS that you've just wrapped up in compiler design
>>7737844
>>7737900
Yeah, I used C++ and OpenCV for this. Generally you can do it in MATLAB/Octave/R/Python or whatever you are comfortable with.
>>7737868
I missed a lot actually. And the bounding boxes' sizes are not very accurate either.
>>7738490
>>7740089
Yeah, I used multiple weak haar-like feature extractors here then applied Adaboost for classification.
It's still very slow because patches including a small part of the motorbikes won't get rejected until it reaches more than half of the classifiers.
>Be a MLfag
>People keep saying that Deep nets have consciousness
>Look at Google's deep dream man, the computers will replace us soon!
>mfw
What kind of tips are there for someone like myself that just found out what Python was this summer and is giving it a shot again next summer? Btw im also a networking and certs baby :3
u script kiddies have a long way to go
>>7741166
I'm basically in love with OpenCV. Before I found it, I was getting high all the time. I was 500 pounds. I was washing my dishes in the bathtub. But then one of my buddies showed me OpenCV and it changed my life.
Anyone into information theory over here? How do you approach these sorts of exercises? I swear my slides are useless
>>7741170
Tfw.
My EE friend and I were talking and he was like "AI is coming anon. Look at Google deepdream. As we introduce randomness into machines they become self-aware"
>niggawhat
>>7742554
Google Deep Dream shows off a cool technique for analyzing neural networks and clearly presents to the audience that different levels of the network seemingly work at different levels of abstraction.
This shows why deeper networks may be useful in an intuitive way.
>>7735961
kek
>>7742585
uh are you telling me that the hype is due to deepdream's educational value? cause i'm sure most people got no insight in ANNs out of the project, and there are countless better ways to educate people on this
>>7742593
The hype is just because it's cool the first time you ever see it (admit it) and most of all because they called it "deep dream" instead of "neural network feature extraction" or something.
>>7742598
yeah nah looks like shitty le psychedelic generator for 420 generation no thanks
>>7736004
It's gold and white but nice try
>>7742593
This is the article that broke the news. It wasn't until after this article went viral that Google created a gallery of pictures. Then after it continued to go viral they made the code accessible to everyone on github.
http://googleresearch.blogspot.ca/2015/06/inceptionism-going-deeper-into-neural.html
>>7742593
Deep dream aside, the reverse image reconstruction technique they used is pretty useful.
It can be done to verify if your network learned to extract the right thing or not. The whole process was purely optimization (using gradient descent, yes) and pretty intuitive. The original paper wasn't even from google, it was from an Oxford's lab.
But then people decided to slap the "Hurr durr dream" name onto those techniques and the pop-sci AI scene blown up, famous people started to worry that AI will dominate us.
Meanwhile it is just a fucking optimization process and a pretty simple one at that.
>>7734467
Writing implementation of C4.5 decision tree for a graph isomorphism filter and random forests.
merry christmas faggots
>>7742585
https://www.youtube.com/watch?v=AgkfIQ4IGaM
This was published 10 days before the Google blog post. Personally I think it's a lot more informative with regards to the things you're saying than Deep Dream, which seems more like a way of getting teens to think Google has their AIs on LSD.
>>7740815
Not really. Sure, part of designing a language is designing a compiler for said language, but not the other way around. People work on compilers for existing languages all the time.
Shouldn't we have some resources to go in the OP for these generals?
>>7743843
http://ml-class.com for ML beginners ?
What I'm working on?
A trivial upper bound for the number of iterations of the Weisfeiler-Lehman-Algorithm (Colour Refinement) for Graph Isomorphism is n-1, where n is the number of vertices of the graph.
Let me get you in on a secret: There are graphs where this upper bound is tight.
S-s-surely this is more important than Babai's new stuff.
chromatic to diatonic conversion and lambdoma
tfw gonna fail my exams
>>7734477
Paper name first author name file:.PDF
Or, just download on your campus.
>>7734549
That called maximum a posteriori learning, mein freude.
>>7734704
More specifically, anon. Two layers in a typical CNN form a bipartite graph. This means you can compactly represent the weights as a matrix. The image is anon projecting the weights to a grayscale image, where each pixel is a normalized weight and each pixel group represents a hidden unit.
>>7740247
Homotopy type theory.
>>7744124
it's `filetype` n00b
>>7740247
Type Theory
Lambda Calculus
Category Theory
Personally I've mostly hit the category theory angle and I've found it really useful and interesting (personally I'm finding the comp sci side of category theory much more interesting than the math side of it even though I'm a pure math major).
>>7734467
Being an indie dev for a RTS using Spring 95.0
Okay, I don't know if this is the right place to ask this but here goes. I was working on a kernel until recently, not Linux or a work project, just a hobby kernel, so don't blame me on not making it that far. I was trying to get protected mode and segmentation working properly. The problem was that when I loaded the kernel I loaded it one megibyte in and when I loaded the code segment register and made the far jump it would jump to an address I didn't own and triple fault. The only way I could get it to work was to either start the code segment at 0 or jump to the address minus 1 megibyte. All I'm asking is is there a better way? anyone here kernel dev? Any tips?
can someone please explain me how do they do it? http://arxiv.org/pdf/1508.06576v2.pdf
what's the difference between a normal ANN and a CNN?
High school student working on neural net to read and evaluate hand written maths/prose. Also, trying to copy that VR mod that works with your desktop to give you 3d windows from an occulus
>>7746406
From a cursory read: they train a CNN against paintings of a particular style. Once training is complete, they provide a new picture as input and pass the input units through the topology. Once they've passed to the hidden nodes, the process is repeated in reverse to generate a new image bases on what the CNN believes the original image 'looks like' to the network. This is called a generative method. Look at restricted Boltzmann machines, or autoencoders for simple examples.
>>7737876
Ouch
Anyone here getting paid (relatively) well to do research?
>>7736044
Interesting picture, and a solid prediction of modern computer security. Anyone know the date?
>>7747979
Gotta pay the bills. Had a ton of loans. I'm hoping to switch to another job soon. Got financially stabilized with the easy money. Been trying to keep up my computer vision skills in the meantime.
I just failed discrete math ama
>>7748405
1 + 0 = ?
>>7736044
That image makes me happy we use the word "cyber" as a prefix and not "computer" nowadays.
completely new to cs wat do
>>7747984
Bump
>>7743841
No, I'm not just talking about the idea of designing a language. Programming Language Theory is a very big area of study *about* programming languages, not just designing them.
www.techradar.com/news/world-of-tech/how-microsoft-beat-google-at-understanding-images-with-machine-learning-1311683
what does /csg/ think about this?
>>7734467
Is this a shitty degree?
Can't decide between shelling out 45k for a Comp Sci degree vs 14k for a MechE degree..
HALP ME PLS DEAR GOD
>>7749789
idk i like it
>>7749789
Mech E. You'll probably learn how to code too.
>>7749809
I fear that I can be replaced by some 18 y.o kiddy who will write script/code or whatever for cheaper and do it just as well though, ya feel? Where as with engineering that's jut impossible.
>>7749815
Yeah there's a few CS/CIS and programming classes in the course outline.It would just mean I'd have to turn down living with my family while I attend college.
>>7749789
Depends on what you're more interested in, really. I'd say go for the MechE, if you love it, because if you're good enough to get a MechE degree, learning to code well enough to land a job shouldn't be too rough if the job market goes south. Most programmers are really total crap at their jobs.
>>7749819
idk my degree was never about coding and i wanna be a researcher
>>7749789
You learn nothing in CS.
>>7749840
i did
I'm interested in machine learning, anybody here have any experience with it? Where should I start?
>>7734467
Getting a job with my worthless CS degree
>>7749840
Great, then you should be able to help me with my Machine Learning exam paper cause you already know everything taught in a CS course!
>>7749913
>>7749914
>>7734547
Thanks anon
>>7749789
If you think you can do it go for mechE. You can usually get the same job as a CS major and 14k<45k
The only reason to go for CS is if you think you can't do the math.
>>7734467
>Category Theory
Good joke. CS students cannot make any use of category theory because they have no mathematical knowledge to apply it. The simplest examples of category theory are motivated by abstract algebra, algebraic topology and algebraic geometry. All of these are unknown to a CS student.
>>7750449
bitch shut up
Can someone explain me
what's backprop
what's a parameter? i sometimes think i get this but i don't think i ever did desu.
what's a (multi-layer) perceptron?
the difference between normalizing with a sigmoid function and other functions that also reduce the domain to 0-1
similarities between bernoulli and gaussian
the curse of probability -- does it only affect when the algorithm uses euclidian distance?
thanks
