The Student Room Group

I may have invented a better machine learning algorithm!

First of all, I don't know whether this algorithm works to generate 'amazing' results like the results produced from current algorithms. But I wish to test it soon. The genre of these 'algorithm' is of course the recurrent neural network model.
The only way I think my algorithm is 'better' is because a RNN takes a while to process the data given (from days to weeks when it comes to industrial work), while my algorithm (if it works) completes its task a lot quicker (maybe under an hour). I'm so confident of this because a RNN uses iterations to process, mine only takes one loop and it's done.

I am about to write this program to test it. However, I also want to document my findings so that if someone recognises that my idea is a good one but the algorithm fails to meet the standards, then they can take my work and improve it to produce something that will indeed work at the end of the day which is better than the RNN, making us move on in the realm of AI.

The way in which I'm doing this, is via YouTube. I will be programming it step by step and testing it, and then uploading it to YouTube to get an audience to judge this whole thing. Before I do that, my question to you is, is there anything I have to do before revealing this to the public? Like any serious procedures I have to undergo to make sure people don't copy my idea etc. The algorithm is simple to the extent where a Year 12 candidate should understand it all while the RNN is a bit more complex. I'm trying to see if my idea has already been explored before. Just to confirm, my idea is not an improvement of the RNN model, it is a different idea as a whole.
Please give me your suggestions. Much appreciated!
(edited 6 years ago)
it will be difficult to keep your idea private property if you put it up on youtube.
if you are serious about keeping this revolutionary discovery private, you should approach a company like Google and give them an overview. they can decide if it is promising for them to develop & will recompense you accordingly
Reply 3
Original post by the bear
it will be difficult to keep your idea private property if you put it up on youtube.


Well, yes, so is there some 'procedure' to still keep it private but I'm able to post it wherever, and if they refer it in their work, my name should be mentioned etc. I don't mean to sound this needy at the infancy of this journey which could go all to waste if it doesn't work but I just want to clear my doubts :smile: .
Original post by fp1washard
Well, yes, so is there some 'procedure' to still keep it private but I'm able to post it wherever, and if they refer it in their work, my name should be mentioned etc. I don't mean to sound this needy at the infancy of this journey which could go all to waste if it doesn't work but I just want to clear my doubts :smile: .


if it is an important discovery then you deserve to reap the benefits. if you put it on the internet then it will be immediately stolen by unscrupulous actors and you will get nothing.
Reply 5
Original post by the bear
if you are serious about keeping this revolutionary discovery private, you should approach a company like Google and give them an overview. they can decide if it is promising for them to develop & will recompense you accordingly


I appreciate this comment, this means I have to test my idea and compare the results to what would be produced by algorithms today before even thinking about that. Hmm. If it does work, I'll hold onto this thought, but it is again a trivial idea.
Reply 6
Original post by the bear
if it is an important discovery then you deserve to reap the benefits. if you put it on the internet then it will be immediately stolen by unscrupulous actors and you will get nothing.


I'm so naive to these things. You may have enlightened me. Many thanks! I'll go and test this, and post my results on here. If I wish to upload anything on YouTube, it may just be the results and not the procedure behind it. Again, it needs to work as good as a RNN does- just quicker.
Original post by fp1washard
I'm so naive to these things. You may have enlightened me. Many thanks! I'll go and test this, and post my results on here. If I wish to upload anything on YouTube, it may just be the results and not the procedure behind it. Again, it needs to work as good as a RNN does- just quicker.


good luck ! i do not have the first idea what all this is about, but i hope it goes well :h:
Reply 8
Original post by the bear
good luck ! i do not have the first idea what all this is about, but i hope it goes well :h:


Music generated by computers :smile:
and thanks! :smile:
Original post by fp1washard
First of all, I don't know whether this algorithm works to generate 'amazing' results like the results produced from current algorithms. But I wish to test it soon. The genre of these 'algorithm' is of course the recurrent neural network model.
The only way I think my algorithm is 'better' is because a RNN takes a while to process the data given (from days to weeks when it comes to industrial work), while my algorithm (if it works) completes its task a lot quicker (maybe under an hour). I'm so confident of this because a RNN uses iterations to process, mine only takes one loop and it's done.

I am about to write this program to test it. However, I also want to document my findings so that if someone recognises that my idea is a good one but the algorithm fails to meet the standards, then they can take my work and improve it to produce something that will indeed work at the end of the day which is better than the RNN, making us move on in the realm of AI.

The way in which I'm doing this, is via YouTube. I will be programming it step by step and testing it, and then uploading it to YouTube to get an audience to judge this whole thing. Before I do that, my question to you is, is there anything I have to do before revealing this to the public? Like any serious procedures I have to undergo to make sure people don't copy my idea etc. The algorithm is simple to the extent where a Year 12 candidate should understand it all while the RNN is a bit more complex. I'm trying to see if my idea has already been explored before. Just to confirm, my idea is not an improvement of the RNN model, it is a different idea as a whole.
Please give me your suggestions. Much appreciated!


You have to decide if your algorithm should be closed source or open source.

If it's closed source and you think it has some commercial value, you should contact the IPO (assuming you are in the UK) to inquire about patenting your algorithm.

https://www.gov.uk/government/organisations/intellectual-property-office

If you want to go open source and disseminate your findings after they have been peer-reviewed, I suggest writing a paper in the IEEE format and submitting it to IEEE for publication:

https://www.ieee.org/publications_standards/publications/authors/publishing_benefits/how_to_publish_with_ieee_presentation.pdf
Reply 10
Original post by jestersnow
You have to decide if your algorithm should be closed source or open source.

If it's closed source and you think it has some commercial value, you should contact the IPO (assuming you are in the UK) to inquire about patenting your algorithm.

https://www.gov.uk/government/organisations/intellectual-property-office

If you want to go open source and disseminate your findings after they have been peer-reviewed, I suggest writing a paper in the IEEE format and submitting it to IEEE for publication:

https://www.ieee.org/publications_standards/publications/authors/publishing_benefits/how_to_publish_with_ieee_presentation.pdf


Ok wow, thank you so much! This is the sort of suggestion I was looking for at the start. :smile:
Reply 11
Original post by the bear
good luck ! i do not have the first idea what all this is about, but i hope it goes well :h:


ok, so I have some results after a few days of testing and building this algorithm. So the aim was to have the computer output sentences it thinks makes sense using sentences I give it to look at. Of course, it's not allow to copy my sentences and output them straight back at me.

I used all the sentences in the first Harry Potter book (since I knew I could find a pdf of it online).
The computer took 20 minutes to analyse all 80000 words and I asked for 500 words for it to output. The computer has distinguished the patterns within the grammar in the English Language like: using full stops and having a capital letter after it etc. You might see numbers, these were the page numbers which I couldn't get rid of when I gave it the book to look at. I also notice that it sometimes opens a speech mark, but never comes to closing it. I could come to solving this at a later date.

Although it cannot judge whether a sentence makes sense, I think it was worthwhile.


1 Harry tried to score," Harry with bulging, lamp and a bit of himself turning teacups into Hagrid's flute to him on alchemy is very good, is Crabbe and found a unicorn blood. The Dursleys had already have much more ferociously as cool logic in his doom. Then he turned to his head, crying, "Harry! Harry, but looking for a low rumbling sound of the ropes binding Harry of the barrier had fallen books. They hadn't received your head, and yet -- speedy -- and out a deserted side of them up two agreed to shut up, boy, and found out of two --" There are you here, Peeves," said Dumbledore, sit," said Peeves -- I keep ter use of neat, straight into the room, his money's worth, just sat down through it in bed for the barrier he said. "The Chasers try takin' risks. Follow me that the towering Christmas ones. "We're looking straight into material. "It's Hedwig!" said Professor Dumbledore around, shook its feet, and it once, and down in the hem of funny-looking people in his seat, but if the scraping noise and their way into a Christmas to breathe on its envelope. "Yer not such a school and got any of a nasty crack as wide. He bought him we could want! The room that they were delighted when two are sure they called Harry. "No one. So if you leave it had to be reaching Hogwarts robes. Harry and he seemed ages ago," said Neville, his new year before now, I am I can you don't think they were perfectly normal, thank you say nothing but looking disgruntled sort of the grin through the silver boats as though this is at his own joke. His face was carrying tumbled out an' more money outta kinda trances. Don~ reckon it's true," Harry often spoke about the book's shrieks and crooked, as they had clicked on her pockets full of us are -- Professor McGonagall -- he was wise, but no matter if it and Ron and speeds toward the boy, we might have noticed that led them right before he yawned and empty seat. No more than they were, he pushed it -- an' that." Harry leaned over Gryffindor meant yes. He tried to Ron, but the lesson, he had grown up at his cupboard -- the storm. "How to find him with her nose and bacon on Malfoy's always the house on her glasses dangling over the best; I don' s'pose it had gotten through a glass or why he left to Dumbledore," said Uncle Vernon rapped on the middle fork, right, Hermione?" "No, we're getting started talking to King's Cross Station. It was a shelf with him feel so they are you?" sneered Uncle Vernon's lap. "He works at the ground!" said Hagrid, I always been to ward off another wizard's robes -- she's really wipe the people off his life. The Quidditch robes up its eyes are not jus' dump him, so they said
(edited 6 years ago)
Original post by fp1washard

Although it cannot judge whether a sentence makes sense, I think it was worthwhile....




very promising :h:

have you read http://www.botnik.org/content/harry-potter.html

?

Quick Reply