Creating an Artificial Minds Eye

You can randomly set up a word vector for each word, but depending on chance some samples of random choices would work better than others, infact with the different input vectors , the loss landscape is altered...... as it is word vectors are not random....so we could imagine making them less random through the learning process by backpropagating a loss that takes in the inputs as variables, rather than the weights of the network...i believe that perhaps when we learn new information, the representations of the words (i.e. how they are understood) is altered fundamentally and the BNN then uses the exact same network to process these differing variables...so when we learn information relating "x is a y" the actual representation of the four words changes...along with a whole lot of other words (perhaps ALL words) in order to give you the disposition to talk about this relationship in different ways without having to train a new network. Eg. if "john is a star student" then the words variables for "john could be the headboy of our school" will all change to reflect this disposition for one to utter the second statement after learning the first...you can imagine that there are countless other statements that must have the variables changed after learning even a simple fact.... It is also my supposition that if we represented words with blobs of colours and placed the words correlating with a sentence in a scene , if we were to train a system with this representation on a conversation corpus, as we alter the representation of the words , the scenery painted by these words will form recognisable (by us) features. Akin to how you can *see* what you are talking about in your mind...in any case the system will change the representation of the words to make it simpler to relate them...so the word "apple", since it produces the disposition to talk about apples, as we change the words that are used to talk about apples, the representation of the apple in the image will also change...so if we say an apple is round, or allude to roundness when talking of an apple, the image with just this information could alter the blob representing the apple multiple ways, so lets say it makes it look rather like a square...then there will be a conflict in how square apples ,say bounce, when we add sentences that refer to bouncing in the training set, and the variable bounce (or how we choose to use the word in conversation) would lead upon learning its dispositions ,to the square shape of the blob to be truly round in our sense. Practically you will need to refer to ALOT of things so that when all these variables constrain each other the apple will become truly round...so. basically the way we use words places restrictions on the nature of the variables we use to represent them and those restrictions can be translated into form....the loss function measures how simple the description the words we paint in the scenery is...if we believe that when we use words our use of words correlate with images of the world, AND that that is the simplest set of images that it can correlate with...then when the Loss is zero we have described just such a world and the images will display it And perhaps mentally ill people are stuck in local minima of such a loss function

Comments

Post a Comment

Popular posts from this blog

ABSTRACT submited

BACKPROPAGATION ON A NOVEL NETWORK (NON ANN) FOR HATESPEECH DETECTION AND MORE This paper introduces a novel type of network that can accurately classify "hate speech" and differentiate it from normal content. It develops a network centered around a unique POS (part of speech) tagging type of system whose specifics are learnt by the network. Each word is assigned variables that indicate the polarity, (sign) of the effect a particular word has on every other word in a sentence, the magnitude of that polarity and also to what extent the word filters itself from the effect from other words. When choosing a set of properties, we would like to properly understand what we want the network to do. We know that both posts classified as hate speech and those not contain samples of words from a common pool (they share words). But we still want that the results for opposite classes to polarize. On top of this I believe that this represents a complex system, in that small chang...
Read more
This involves a "hate speech" or "spam" detection algorithm. (depending on the training data). This will be based on ANN theory , and involve the backpropagation algorithm while not based on an actual neural network. The training data will be labeled into two classes, "hate speech" and "non hate speech".   Hate speech will have an expected value of 1 , representing certain hate speech, while non hate speech will have an expected value of -1.   So what we would like is for the software to be input a post and gives it a value of 1 or -1, or something much closer to one or the other. During training The program will compute a value by performing the equivalent of forward propagation through the parameters we assign to each of multiple structures we will embed in the input ...then using this value and the epected value we will compute a cost function. We will then backpropagate through the heirachical structure of parameters in the direc...
Read more