Neural Networks are a class of models within the general machine learning literature. Neural networks are a specific set of algorithms that has revolutionized the field of machine learning. They are inspired by biological neural networks and the current so called deep neural networks have proven to work quite very well. Neural Networks are themselves general function approximations, that is why they can be applied to literally almost any machine learning problem where the problem is about learning a complex mapping from the input to the output space.
Top 10 Neural Network Architectures You Need to Know

1 — Perceptrons

Considered the first generation of neural networks, Perceptrons are simply computational models of a single neuron. Perceptron was originally coined by Frank Rosenblatt in his paper, The perceptron: a probabilistic model for information storage and organization in the brain” (1956) [1]. Also called feed-forward neural network, perceptron feeds information from the front to the back. Training perceptrons usually requires back-propagation, giving the network paired datasets of inputs and outputs. Inputs are sent into the neuron, processed, and result in an output. The error being back propagated is often some variation of the difference between the input and the output. Given that the network has enough hidden neurons, it can theoretically always model the relationship between the input and output. Practically their use is a lot more limited but they are popularly combined with other networks to form new networks.
However, Perceptrons do have limitations: If you choose features by hand and you have enough features, you can do almost anything. For binary input vectors, we can have a separate feature unit for each of the exponentially many binary vectors and so we can make any possible discrimination on binary input vectors. But once the hand-coded features have been determined, there are very strong limitations on what a perceptron can learn.
2 — Convolutional Neural Networks
In 1998, Yann LeCun and his collaborators developed a really good recognizer for handwritten digits called LeNet. It used back propagation in a feedforward net with many hidden layers, many maps of replicated units in each layer, pooling of the outputs of nearby replicated units, a wide net that can cope with several characters at once even if they overlap, and a clever way of training a complete system, not just a recognizer. Later it is formalized under the name convolutional neural networks (CNNs). Fun fact: This net was used for reading ~10% of the checks in North America.
Convolutional Neural Networks are quite different from most other networks. They are primarily used for image processing but can also be used for other types of input such as as audio. A typical use case for CNNs is where you feed the network images and the network classifies the data. CNNs tend to start with an input “scanner” which is not intended to parse all the training data at once. For example, to input an image of 100 x 100 pixels, you wouldn’t want a layer with 10 000 nodes. Rather, you create a scanning input layer of say 10 x 10 which you feed the first 10 x 10 pixels of the image. Once you passed that input, you feed it the next 10 x 10 pixels by moving the scanner one pixel to the right.
This input data is then fed through convolutional layers instead of normal layers, where not all nodes are connected to all nodes. Each node only concerns itself with close neighboring cells. These convolutional layers also tend to shrink as they become deeper, mostly by easily divisible factors of the input. Besides these convolutional layers, they also often feature pooling layers. Pooling is a way to filter out details: a commonly found pooling technique is max pooling, where we take say 2 x 2 pixels and pass on the pixel with the most amount of red.

3 — Recurrent Neural Networks
To understand RNNs, we need to have a brief overview on sequence modeling. When applying machine learning to sequences, we often want to turn an input sequence into an output sequence that lives in a different domain; for example, turn a sequence of sound pressures into a sequence of word identities. When there is no separate target sequence, we can get a teaching signal by trying to predict the next term in the input sequence. The target output sequence is the input sequence with an advance of 1 step. This seems much more natural than trying to predict one pixel in an image from the other pixels, or one patch of an image from the rest of the image. Predicting the next term in a sequence blurs the distinction between supervised and unsupervised learning. It uses methods designed for supervised learning, but it doesn’t require a separate teaching signal.

Memoryless models are the standard approach to this task. In particular, autoregressive models can predict the next term in a sequence from a fixed number of previous terms using “delay taps; and feed-forwad neural nets are generalized autoregressive models that use one or more layers of non-linear hidden units. However, if we give our generative model some hidden state, and if we give this hidden state its own internal dynamics, we get a much more interesting kind of model: It can store information in its hidden state for a long time. If the dynamics is noisy and the way it generates outputs from its hidden state is noisy, we can never know its exact hidden state. The best we can do is to infer a probability distribution over the space of hidden state vectors. This inference is only tractable for 2 types of hidden state model.
Originally introduced in Jeffrey Elman’s Finding structure in time” (1990) [3], recurrent neural networks (RNNs) are basically perceptrons; however, unlike perceptrons which are stateless, they have connections between passes, connections through time. RNNs are very powerful, because they combine 2 properties: 1) distributed hidden state that allows them to store a lot of information about the past efficiently; and 2) non-linear dynamics that allows them to update their hidden state in complicated ways. With enough neurons and time, RNNs can compute anything that can be computed by your computer. So what kinds of behavior can RNNs exhibit? They can oscillate, they can settle to point attractors, they can behave chaotically. And they could potentially learn to implement lots of small programs that each capture a nugget of knowledge and run in parallel, interacting to produce very complicated effects.

One big problem with RNNs is the vanishing (or exploding) gradient problem where, depending on the activation functions used, information rapidly gets lost over time. Intuitively this wouldn’t be much of a problem because these are just weights and not neuron states, but the weights through time is actually where the information from the past is stored; if the weight reaches a value of 0 or 1 000 000, the previous state won’t be very informative. RNNs can in principle be used in many fields as most forms of data that don’t actually have a timeline (i.e. unlike sound or video) can be represented as a sequence. A picture or a string of text can be fed one pixel or character at a time, so the time dependent weights are used for what came before in the sequence, not actually from what happened x seconds before. In general, recurrent networks are a good choice for advancing or completing information, such as autocompletion.
4 — Long / Short Term Memory
Hochreiter & Schmidhuber (1997) [4] solved the problem of getting a RNN to remember things for a long time by building what known as long-short term memory networks (LSTMs). LSTMs networks try to combat the vanishing / exploding gradient problem by introducing gates and an explicitly defined memory cell. The memory cell stores the previous values and holds onto it unless a “forget gate” tells the cell to forget those values. LSTMs also have a “input gate” which adds new stuff to the cell and an “output gate” which decides when to pass along the vectors from the cell to the next hidden state.

Recall that with all RNNs, the values coming in from X_train and H_previous are used to determine what happens in the current hidden state. And the results of the current hidden state (H_current) are used to determine what happens in the next hidden state. LSTMs simply add a cell layer to make sure the transfer of hidden state information from one iteration to the next is reasonably high. Put another way, we want to remember stuff from previous iterations for as long as needed, and the cells in LSTMs allow this to happen. LSTMs have been shown to be able to learn complex sequences, such as writing like Shakespeare or composing primitive music.
5 — Gated Recurrent Unit
Gated recurrent units (GRUs) are a slight variation on LSTMs. They take X_train and H_previous as inputs. They perform some calculations and then pass along H_current. In the next iteration and H_current are used for more calculations, and so on. What makes them different from LSTMs is that GRUs don’t need the cell layer to pass values along. The calculations within each iteration insure that the H_current values being passed along either retain a high amount of old information or are jump-started with a high amount of new information.

In most cases, GRUs function very similarly to LSTMs, with the biggest difference being that GRUs are slightly faster and easier to run (but also slightly less expressive). In practice these tend to cancel each other out, as you need a bigger network to regain some expressiveness which then in turn cancels out the performance benefits. In some cases where the extra expressiveness is not needed, GRUs can outperform LSTMs. You can read more about GRU from Junyoung Chung’s 2014 paper Empirical evaluation of gated recurrent neural networks on sequence modeling [5].
6 — Hopfield Network
Recurrent networks of non-linear units are generally very hard to analyze. They can behave in many different ways: settle to a stable state, oscillate, or follow chaotic trajectories that cannot be predicted far into the future. To resolve this problem, John Hopfield introduced Hopfield Net in his 1982 paper Neural networks and physical systems with emergent collective computational abilities [6]. A Hopfield network (HN) is a network where every neuron is connected to every other neuron; it is a completely entangled plate of spaghetti as even all the nodes function as everything. Each node is input before training, then hidden during training and output afterwards. The networks are trained by setting the value of the neurons to the desired pattern after which the weights can be computed. The weights do not change after this. Once trained for one or more patterns, the network will always converge to one of the learned patterns because the network is only stable in those states.

There is another computational role for Hopfield nets. Instead of using the net to store memories, we use it to construct interpretations of sensory input. The input is represented by the visible units, the interpretation is represented by the states of the hidden units, and the badness of the interpretation is represented by the energy.
Unfortunately people shown that Hopfield net is very limited in its capacity. A Hopfield net of N units can only memorize 0.15N patterns because of the so-called spurious minima in its energy function. The idea is that since the energy function is continuous in the space of its weights, if two local minima are too close, they might “fall” into each other to create a single local minima which doesn’t correspond to any training sample, while forgetting about the two samples it is supposed to memorize. This phenomenon significantly limits the number of samples that a Hopfield net can learn.
7 — Boltzmann Machine
A Boltzmann Machine is a type of stochastic recurrent neural network. It can be seen as the stochastic, generative counterpart of Hopfield nets. It was one of the first neural networks capable of learning internal representations, and is able to represent and solve difficult combinatoric problems. First introduced by Geoffrey Hinton and Terrence Sejnowski in Learning and relearning in Boltzmann machines” (1986) [7], Boltzmann machines are a lot like Hopfield Networks, but some neurons are marked as input neurons and others remain “hidden”. The input neurons become output neurons at the end of a full network update. It starts with random weights and learns through back-propagation. Compared to a Hopfield Net, the neurons mostly have binary activation patterns.
The goal of learning for Boltzmann machine learning algorithm is to maximize the product of the probabilities that the Boltzmann machine assigns to the binary vectors in the training set. This is equivalent to maximizing the sum of the log probabilities that the Boltzmann machine assigns to the training vectors. It is also equivalent to maximizing the probability that we would obtain exactly the N training cases if we did the following: 1) Let the network settle to its stationary distribution N different time with no external input; and 2) Sample the visible vector once each time.

An efficient mini-batch learning procedure was proposed for Boltzmann Machines by Salakhutdinov and Hinton in 2012 [8].

  • For the positive phase, first initialize the hidden probabilities at 0.5, then clamp a data vector on the visible units, then update all the hidden units in parallel until convergence using mean field updates. After the net has converged, record PiPj for every connected pair of units and average this over all data in the mini-batch.

  • For the negative phase: first keep a set of “fantasy particles.” Each particle has a value that is a global configuration. Then sequentially update all the units in each fantasy particle a few times. For every connected pair of units, average SiSj over all the fantasy particles.

In a general Boltzmann machine, the stochastic updates of units need to be sequential. There is a special architecture that allows alternating parallel updates which are much more efficient (no connections within a layer, no skip-layer connections). This mini-batch procedure makes the updates of the Boltzmann machine more parallel. This is called a Deep Boltzmann Machine (DBM), a general Boltzmann machine with a lot of missing connections.
8 — Deep Belief Networks
Back-propagation is considered the standard method in artificial neural networks to calculate the error contribution of each neuron after a batch of data is processed. However, there are some major problems using back-propagation. Firstly, it requires labeled training data; while almost all data is unlabeled. Secondly, the learning time does not scale well, which means it is very slow in networks with multiple hidden layers. Thirdly, it can get stuck in poor local optima, so for deep nets they are far from optimal.
To overcome the limitations of back-propagation, researchers have considered using unsupervised learning approaches. This helps keep the efficiency and simplicity of using a gradient method for adjusting the weights, but also use it for modeling the structure of the sensory input. In particular, they adjust the weights to maximize the probability that a generative model would have generated the sensory input. The question is what kind of generative model should we learn? Can it be an energy-based model like a Boltzmann machine? Or a causal model made of idealized neurons? Or a hybrid of the two?

Yoshua Bengio came up with Deep Belief Networks in his 2007 paper Greedy layer-wise training of deep networks [9], which have been shown to be effectively trainable stack by stack. This technique is also known as greedy training, where greedy means making locally optimal solutions to get to a decent but possibly not optimal answer. A belief net is a directed acyclic graph composed of stochastic variables. Using belief net, we get to observe some of the variables and we would like to solve 2 problems: 1) The inference problem: Infer the states of the unobserved variables, and 2) The learning problem: Adjust the interactions between variables to make the network more likely to generate the training data.
Deep Belief Networks can be trained through contrastive divergence or back-propagation and learn to represent the data as a probabilistic model. Once trained or converged to a stable state through unsupervised learning, the model can be used to generate new data. If trained with contrastive divergence, it can even classify existing data because the neurons have been taught to look for different features.
9 — Autoencoders
Autoencoders are neural networks designed for unsupervised learning, i.e. when the data is not labeled. As a data-compression model, they can be used to encode a given input into a representation of smaller dimension. A decoder can then be used to reconstruct the input back from the encoded version.
The work they do is very similar to Principal Component Analysis, which is generally used to represent a given input using less number of dimensions than originally present. So for example, in NLP if you represent a word as a vector of 100 numbers, you could use PCA to represent it in 10 numbers. Of course, that would result in loss of some information, but it is a good way to represent your input if you can only work with a limited number of dimensions. Also, it is a good way to visualize the data because you can easily plot the reduced dimensions on a 2D graph, as opposed to a 100-dimensional vector. Autoencoders do similar work — the difference being that they can use non-linear transformations to encode the given vector into smaller dimensions (as compared to PCA which is a linear transformation). So it can generate more complex encodings.

They can be used for dimension reduction, pretraining of other neural networks, for data generation etc. There are a couple of reasons: (1) They provide flexible mappings both ways, (2) the learning time is linear (or better) in the number of training cases, and (3) the final encoding model is fairly compact and fast. However, it turned out to be very difficult to optimize deep auto encoders using back propagation. With small initial weights, the back propagated gradient dies. Nowadays they are rarely used in practical applications, mostly because in key areas for which they where once considered to be a breakthrough (such as layer-wise pre-training), it turned out that vanilla supervised learning works better. Check out the original paper by Bourlard and Kamp dated back in 1988 [10].
10 — Generative Adversarial Network
In Generative adversarial nets” (2014) [11], Ian Goodfellow introduced a new breed of neural network, in which 2 networks work together. Generative Adversarial Networks (GANs) consist of any two networks (although often a combination of Feed Forwards and Convolutional Neural Nets), with one tasked to generate content (generative) and the other has to judge content (discriminative). The discriminative model has the task of determining whether a given image looks natural (an image from the dataset) or looks like it has been artificially created. The task of the generator is to create natural looking images that are similar to the original data distribution. This can be thought of as a zero-sum or minimax two player game. The analogy used in the paper is that the generative model is like “a team of counterfeiters, trying to produce and use fake currency” while the discriminative model is like “the police, trying to detect the counterfeit currency”. The generator is trying to fool the discriminator while the discriminator is trying to not get fooled by the generator. As the models train through alternating optimization, both methods are improved until a point where the “counterfeits are indistinguishable from the genuine articles”.

According to Yann LeCun, these networks could be the next big development. They are one of the few successful techniques in unsupervised machine learning, and are quickly revolutionizing our ability to perform generative tasks. Over the last few years, we’ve come across some very impressive results. There is a lot of active research in the field to apply GANs for language tasks, to improve their stability and ease of training, and so on. They are already being applied in industry for a variety of applications ranging from interactive image editing, 3D shape estimation, drug discovery, semi-supervised learning to robotics.

Neural networks are one of the most beautiful programming paradigms ever invented. In the conventional approach to programming, we tell the computer what to do, breaking big problems up into many small, precisely defined tasks that the computer can easily perform. By contrast, in a neural network we don’t tell the computer how to solve our problem. Instead, it learns from observational data, figuring out its own solution to the problem at hand.
Today, deep neural networks and deep learning achieve outstanding performance on many important problems in computer vision, speech recognition, and natural language processing. They’re being deployed on a large scale by companies such as Google, Microsoft, and Facebook. I hope that this post helps you learn the core concepts of neural networks, including modern techniques for deep learning.