Scribd
Upload
23 views3 pages

Background

The document discusses various natural language processing and computer vision topics including word embeddings, sentiment classification, and face recognition. Word embeddings assign numeric representations to words to capture semantic meaning. Sentiment classification involves building and training a recurrent neural network model to classify movie reviews as positive or negative. Face recognition aims to identify people from images by training convolutional neural networks, but requires retraining for new individuals.

Uploaded by

Gajanan Tale
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
23 views3 pages

Background

The document discusses various natural language processing and computer vision topics including word embeddings, sentiment classification, and face recognition. Word embeddings assign numeric representations to words to capture semantic meaning. Sentiment classification involves building and training a recurrent neural network model to classify movie reviews as positive or negative. Face recognition aims to identify people from images by training convolutional neural networks, but requires retraining for new individuals.

Uploaded by

Gajanan Tale
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 3

Background

 Before you begin with this course, it is assumed you have completed the
course Convolutional Neural Networks.
 You are now familiar with the concepts on convolution, striding, padding, pooling,
and object classification.
 Going forward you will be using these concepts not only to classify different
objects in the same image but also to locate them.


1  of 4

SimilarWord to Vec
 Learning good word embeddings are of paramount importance in NLP.
 A good word embedding is one which can represent a word in minimum vector
space and at the same time preserve the semantics as well as their context in
the language.
 Each word embeddings are points in a vector space, and the transformation of
words to their vector representation is called wordtovec.

ity Function
 One shot learning is based on similarity function.
 Consider, we have two images img1img1 an

Word Embeddings
 The primary objective of Natural Language Processing especially in deep
learning is to attain human-like performance in a linguistic task.
 For a machine, a word is a just a string, and it does not make any difference if
the words are used in different context.
 For example, humans can understand that words one, two, three.. falls under
one category when compared to words I, we, you ..
 Hence we need an algorithm that can assign a numeric representation for words
such that words with similar context will have similar numeric representation.
 These numeric representations are known as word embeddings and in the rest of
the course, you will explore different types of word embedding and the algorithm
behind them.
 One naive way of representing a word in numeric format is by one-hot encoding.
 As shown in the image initially, all the words in the vocabulary are stored as a set
and assigned a unique index to each word.
 Later each word is represented as a vector, where all the elements are zeros
except for the index of the word which is equal to one.

FacFace Recognition
 A face recognition system should be able to recognize a person's identity
provided that it has been trained to identify that person among other people
uniquely.
 One way to build this system is to train a CNN network with many images of the
same person.

The flaw Learning Word Embeddings


 There are several ways of arriving at word vectors using deep learning and some
of the popular methods are

1. Continuous Bag of words


2. Skipgram model
3. Glovec model

In the rest of the cards, in this topic, let's try to understand the theoretical
implementation of these models and understand practical implementation in the next
topic.

Sentiment Classification
 This topic is the road map for the next hands-on, where you train a model to
classify movie reviews to predict if a review is a positive review or a negative
review.
 You will perform following steps to build and train the classifier.
o Perform necessary preprocessing to transform movie reviews into one hot
encoded data.
o Initialize the lookup table.
o Build an RNN model to learn from the input data.
o Use the output of RNN to classify the reviews using sigmoid activation.
 You will be using Keras framework to perform all the above operations.
 of this approach is that we have limited training images and every time if there is
a new person, the network has to be retrained.

e Recognition
 A face recognition system should be able to recognize a person's identity
provided that it has been trained to identify that person among other people
uniquely.
 One way to build this system is to train a CNN network with many images of the
same person.
 The flaw of this approach is that we have limited training images and every time if
there is a new person, the network has to be retrained.

Face Recognition
 A face recognition system should be able to recognize a person's identity
provided that it has been trained to identify that person among other people
uniquely.
 One way to build this system is to train a CNN network with many images of the
same person.
 The flaw of this approach is that we have limited training images and every time if
there is a new person, the network has to be retrained.

You might also like