TensorFlow vs Google AI & # 8211; Deep dream of Google
When the "deep learning" storms continue, Google's AIs are constantly evolving and Google's search engine understands everything on the web. Google's AIs include deep dream, chatbot, spam-killing AI and image recognition. In order to develop smart AI, Google cannot lack "massive" libraries.
TensorFlow was developed by Google and released in October 2015. This library supports the construction of very complex deeplearning models through extremely short APIs. Tensor-based deeplearning models can be used on many different platform types (from smartphones to distributed servers) and on both CPUs and GPUs.
DeepDream was developed by Google to identify and enhance the pattern of photographs through pareidolia psychedelic algorithms from which to create images like dreams. DeepDream is built on deeplearning technology with convolutional neural network model – the most advanced technology for image recognition and applied in AlphaGo and many other modern AI games. Google's image recognition tool is also built on this technology.
Convolutional model neural network is one of the built-in models in TensorFlow. With just a few short setup steps, users were able to build a convolutional model of neural network to use.
Code in TensorFlow:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 | def conv2d (x, W): return tf.nn.conv2d (x, W, strides = [1, 1, 1, 1], padding = 'SAME') def max_pool_2x2 (x): return tf.nn.max_pool (x, ksize = [1, 2, 2, 1], strides = [1, 2, 2, 1], padding = 'SAME') # Layer 1 W_conv1 = weight_variable ([5, 5, 1, 32]) b_conv1 = bias_variable ([32]) h_conv1 = tf.nn.relu (conv2d (x_image, W_conv1) + b_conv1) h_pool1 = max_pool_2x2 (h_conv1) # Layer 2 W_conv2 = weight_variable ([5, 5, 32, 64]) b_conv2 = bias_variable ([64]) h_conv2 = tf.nn.relu (conv2d (h_pool1, W_conv2) + b_conv2) h_pool2 = max_pool_2x2 (h_conv2) # Classification layer W_fc1 = weight_variable ([7 * 7 * 64, 1024]) b_fc1 = bias_variable ([1024]) h_pool2_flat = tf.reshape (h_pool2, [-1, 7 * 7 * 64]) h_fc1 = tf.nn.relu (tf.matmul (h_pool2_flat, W_fc1) + b_fc1) |
Google chatbot is developed based on deeplearning technology with model sequence2sequence that can talk to people in an interesting way. This deeplearning model is not only used in chatbot but also for smart translator tool. The seq2seq model is essentially an extension of the recurrent neural network (rnn) model with each model corresponding to an encoder and a decoder. The most interesting thing in the model rnn and seq2seq is that each word is "born" in series and the same way a person talks. More specifically, with the rnn model, a chatbot can compose a novel in a certain style like Sheakspeare's style.
Code in TensorFlow:
1 2 3 4 5 6 7 8 9 | # The seq2seq function: we use embedding for the input and attention. def seq2seq_f (encoder_inputs, decoder_inputs, do_decode): return tf.nn.seq2seq.embedding_attention_seq2seq ( encoder_inputs, decoder_inputs, cell, num_encoder_symbols = source_vocab_size, num_decoder_symbols = target_vocab_size, embedding_size = size, output_projection = output_projection, feed_previous = do_decode) |
The cnn, rnn and seq2seq models are used for both text, images and other data types.
TensorFlow is an open source library useful for smart applications using the most advanced deeplearning technologies available today.
ITZone via Techmaster