I'm guessing it's related to using the sigmoid function on the output but I'd like to have a proper understanding of the math behind it. Since each edge represents the computation of one chain rule, connecting some node to one of its parent nodes. For learning, we want to find the gradient of the cost function. And the last bit of extension, if one of the input values, for example x is also dependent on it’s own inputs. The problem l ies in the implementation of the Backpropagation algorithm itself. Linear Algebra with Applications. ALGORITHM 1. It is the method of fine-tuning the weights of a neural net based on the error rate obtained in the previous epoch (i.e., iteration). 2). For the table of contents and more content click here. The gradient is a vector of slopes for a function along multiple axes. In memoization we store previously computed results to avoid recalculating the same function. The purpose of learning is to determine the weights W ij that allow us to reproduce the provided patterns of inputs and outputs (function of inputs). This method has the advantage of being readily adaptable to … Maybe improve it a bit. What is learning rate in backpropagation? Therefore, it’s necessary before running the neural network on training data to check if our implementation of backpropagation … Here we start to depart from theory and go into the practical arena. We change the parameters using optimization algorithms. The conjugate gradient algorithms and resilient backpropagation all provide fast convergence. The Backpropagation algorithm is used to learn the weights of a multilayer neural network with ... of backpropagation that seems biologically plausible. TensorFlow is an open source software library for numerical computation using data-flow graphs. In the basic BP algorithm the weights are adjusted in the steepest descent direction (negative of the gradient). When a small change in x produces a large change in the function f, we say the the function is very sensitive to x. When the neural network is initialized, weights are set for its individual elements, called neurons. What is classification by backpropagation? TensorFlow is cross-platform. One popular method was to perturb (adjust) the weights in a random, uninformed direction (ie. The function f can have different sensitivities to each input. The sensitivity is denoted by: To extend this further, let’s say our function was multi-variable now. There is no pure backpropagation or pure feed-forward neural network. This answer is the absolute best explanation, broken down into plain English step by step, that I have found. Taking the derivative of Eq. Neural networks and back-propagation explained in a simple way. The intuition behind this term is that initialization to ⊤ allows the algorithm to propagate information into a cyclic region, optimistically assuming that the value along the back edge will confirm this initial propagation. Since you talk about training until you "reach input level", I assume you train until output is exactly as the target value in the data set. Let’s see how we would get the computational graph for a²₁ through a¹₁. The network is initialized with randomly chosen weights. (3.4) and (3.5) we used, the smaller the changes to the weights and biases of the network will be in one iteration, as well as the smoother the trajectories in the weight and bias space will be. It uses the gradient produced by the back propagation algorithm. It adopts the gradient descent algorithm. Expert Answer 100% (1 rating) The following are true regarding back propagation rule: It is also called generalized delta rule Erro view the full answer. Next post => http likes 246. Then we move on to the preceding 3 computations. Memoization is a computer science term which simply means: don’t recompute the same thing over and over. This is the function applied to often one data point to find the delta between the predicted point and the actual point for example. ... During training, the objective is to reduce the loss function on the training dataset as much as possible. Use gradient descent or a built-in optimization function to minimize the cost function with the weights in theta. This minimization algorithm calculates the gradient of the cost as a function of the weights. But when an analytical method fails or is difficult, we usually try numerical differentiation. I’ll start with a simple one-path network, and then move on to a network with multiple units per layer. It becomes more useful to think of it as a separate thing when you have multiple layers, as unlike your example where you apply the chain rule once, you do need to apply it multiple times, and it is most convenient to apply it layer-by-layer in reverse order to the feed forward steps. Learning algorithm can refer to this Wikipedia page.. In other words, we need to know what effect changing each of the weights will have on E 2. GRADIENT Whereas a derivative or differential is the rate of change along one axis. Rather they are discrete nodes that approximate a function in concert. It was introduced by Naftali Tishby, Fernando C. Pereira, and William Bialek. What is the objective of the backpropagation algorithm? BACK PROPAGATION ALGORITHM. © AskingLot.com LTD 2021 All Rights Reserved. Back-propagation is a method to calculate that gradient. It is the technique still used to train large deep learning networks. In all optimization problems, the objective is to find the maximum or minimum value of a given function with or without constraints. I consider them very different types of algorithms, LM beeing a general non-linear least-squares optimization method, Backpropagation a method for computing gradients of a loss-function in regards to some parameters (it still needs an optimization algorithm). Backpropagation is algorithm to train (adjust weight) of neural network. Under the Hood of K-Nearest Neighbors (KNN) and Popular Model Validation Techniques, How To: Deploy GPT2 NLG with Flask on AWS ElasticBeanstalk, [Paper] NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications (Image…, Introducing Objectron: The Next Phase in 3D Object Understanding, An Introduction to Online Machine Learning, Detecting Breast Cancer using Machine Learning. It employs gradient descent to minimize the loss function between the network outputs and the target values for these outputs. We can keep doing this for arbitrary number of layers. To be continued… Deep Learning with Python and Keras. MIT Press. Sutton, R. S. (2018). Proper tuning of the weights allows you to reduce error rates and to make the model reliable by increasing its generalization. Notice the pattern in the derivative equations below. The Adam optimization algorithm is an extension to stochastic gradient descent that has recently seen broader adoption for deep learning applications in computer vision and natural language processing. Machine Learning FAQ Can you give a visual explanation for the back propagation algorithm for neural networks? popular learning method capable of handling such large learning problems — the backpropagation algorithm. Back-propagation is such an algorithm that performs a gradient descent minimisation of E 2. With this example we have 3 nodes and 2 links. In the derivation of the backpropagation algorithm below we use the sigmoid function, largely because its derivative has some nice properties. Explanation: The objective of backpropagation algorithm is to to develop learning algorithm for multilayer feedforward neural network, so that network … That's a short and broad question. Hence the need for a recursive algorithm to find it’s derivative or gradient, which takes into factor all the nodes. • To understand the role and action of the logistic activation function which is used as a basis for many neurons, especially in the backpropagation algorithm. Now let’s see how we would get the computational graph for a²₂ through a¹₁. itly approximate the backpropagation algorithm (O’Reilly, 1998; Lillicrap, Cownden,Tweed,&Akerman,2016;Balduzzi,Vanchinathan,&Buhmann, 2014; Bengio, 2014; Bengio, Lee, Bornschein, & Lin, 2015; Scellier & Bengio, 2016), and we will compare them in detail in section 4. If you remember DEFINITIONS 6 & 7, specifically 7, you’ll remember that the cost function is conceptually the average or the weighted average of the differences between the predicted and actual outputs. We introduce this concept to illustrate the complicated flow of computations in the back-prop algorithm. FURTHER COMPLICATIONS WITH A COMPLEX MODEL. The backpropagation algorithm learns the weights of a given network. Numerous studies have compared … ... we cover eyes this time so that we can't see where we are and when we accomplished our "objective," that is, reaching the top of the mountain. And if a small change in x produces a small change in f, we say it’s not very sensitive. However, brain connections appear to be unidirectional and not bidirectional as would be required to implement backpropagation. I'm not sure what the purpose of the o(1-o) in the back propagation algorithm achieves? Backpropagation refers to the method of calculating the gradient of neural network parameters. In going forward through the neural net, we end up with a predicted value, a. Backpropagation is an algorithm commonly used to train neural networks. So in this sense we are propagating backwards through the neural network and updating each layer. Where y is the actual value and a is the predicted value. Meaning that if a computation has already been computed, then it could be reused the next and the next time and so on. CONCEPT 2. Backprobagation can be viewed as an optimization problem, as it tries to minimize the cost function between the hypothesis outputs and the actual outputs. If you consider all the nodes in a neural network and the edges that connect them, you can think of the computation required to do back propagation increasing linearly with the number of edges. 1234 J. Whittington and R. Bogacz contrast, for the other output node y(0) 2, there is no path leading to it from the active input node via strong connections, so its activity is low. The backpropagation training algorithm is based on the principle of gradient descent and is given as … Given a function f, we wanna find the gradient: where x is a set of variables whose derivatives we need, and y are additional variables, that we don’t require the derivatives. Remember from earlier, when we defined loss function to be a difference squared, that’s what we use here on the last layer of the computation graph. It runs on nearly everything: GPUs and CPUs—including mobile and embedded platforms—and even tensor processing units (TPUs), which are specialized hardware to do tensor math on. The algorithm should adjust the weights such that E 2 is minimised. The RP algorithm works well on all the pattern recognition problems. One such tool which has demonstrated promising potential is the artificial neural network. 3). Explanation: No feedback is involved at any stage as it is a feedforward neural network. Most times this is the squared loss, which gives the distance measure. For example: For learning, we want to find the gradient of the cost function. Here we aim to build a concrete understanding of the backprop algorithm while still keeping certain complications out of sight. Starting nodes are what you will see in the equation, for the sake of the diagram, there’s always a need to define additional variables for intermediate nodes, in this example the node “u”. What is internal and external criticism of historical sources? Once, the forward propagation is done, the model has to back-propagate and update the weights. Specifically, explanation of the backpropagation algorithm was skipped. The Backpropagation Algorithm Pandamatak May 7th, 2018 - We are now in a position to state the Backpropagation algorithm formally Formal statement of the algorithm Stochastic Backpropagation training examples n i n h n o' We consider the make up of x, and how its ingredients may be affecting the overall effectiveness of the drug. Then we move on to the preceding computation. During the training stage, the input gets carried forward and at the end produces a scalar cost J(θ). Since I have been really struggling to find an explanation of the backpropagation algorithm that I genuinely liked, I have decided to write this blogpost on the backpropagation algorithm for word2vec.My objective is to explain the essence of the backpropagation algorithm using a simple - yet nontrivial - neural network. Doing it analytically in terms of algebra is probably what you did in school. Which algorithm is best depends on the purpose of using an ANN. The objective of this algorithm is to create a training mechanism for neural networks to ensure that the network is trained to map the inputs to their appropriate outputs. The node “u” is equivalent to “mx”. Inputs are loaded, they are passed through the network of neurons, and the network provides an output for each one, given the initial weights. Input consists of several groups of multi-dimensional data set, The data were cut into three parts (each number roughly equal to the same group), 2/3 of the data given to training function, and the remaining 1/3 of the data given to testing function. 4. In the following, we briefly present the algorithm and derive the … So for example, maybe just quantity analysis wasn’t enough, so we break down the drug into 3 active ingredients and consider each one’s dosage. The back-prop algorithm then goes back into the network and adjusts the weights to compute the gradient. Since it’s … Using Java Swing to implement backpropagation neural network. To be continued…. This value that we get from the summation of all preceding nodes and their gradients has the instruction for updating it so that we minimize the error. where, ∂y/∂x is the n×m Jacobian matrix of g. DEFINITION 10. The backpropagation (BP) algorithm using the generalized delta rule (GDR) for gradient calculation (Werbos, Ph.D. Thesis, Harvard University, 1974), has been popularized as a method of training ANNs. The backprop algorithm visits each node only once to calculate the partials, this prevents the unnecessary recalculation of exponential number of sub expressions. (Choose all that apply) 5. (n.d.). The algebraic expression or the computational graph don’t deal with numbers, rather they just give us the theoretical background to verify that we are computing them correctly. While this increases the use of memory, it significantly reduces compute time, and for a large neural net, is necessary. STOCHASTIC GRADIENT DESCENT. Backpropagation. Numerical differentiation is done using discrete points of a function. Since algebraic manipulation is difficult or not possible, with numerical methods we general use methods that are heavy in calculation, therefore computers are often used. From here there are 2 general methods: one is using the nearby points, while the other is using curve fitting. Again with the same example, maybe the x is broken down into it’s constituent parts in the body, so we have to consider that as well. Let’s assume we are really into mountain climbing, and to add a little extra challenge, we cover eyes this time so that we can’t see where we are and when we accomplished our “objective,” that is, reaching the top of the mountain. Prentice-Hall. Also, I’ve mentioned it is a somewhat complicated algorithm and that it deserves the whole separate blog post. A Visual Explanation of the Back Propagation Algorithm for Neural Networks = Previous post. A Bradford Book. These classes of algorithms are all referred to generically as "backpropagation". Feedforward Networks: Nomenclature Consider a feedforward network f W Rn! Then for Neural Networks we use the Back Propagation algorithm. Why? Examples: Deriving the base rules of backpropagation In machine learning, backpropagation (backprop, BP) is a widely used algorithm for training feedforward neural networks. COMPLICATIONS WITH A COMPLEX MODEL. We order them in such a way that we the computation of one comes after the other. To expand it to realistic networks, like this. The backpropagation algorithm is key to supervised learning of deep neural networks and has enabled the recent surge in popularity of deep learning algorithms since … • To study and derive the backpropagation algorithm. The gradient of a value z with respect to this tensor is. First of all we have to make a few setups, first of those being, the order of the neural network and the computational graph of the nodes associated with our network. • To study and derive the backpropagation algorithm. N, one with n real inputs and N output units. The backpropagation (BP) algorithm that was introduced by Rumelhart [6] is the well-known method for training a multilayer feed-forward artificial neural networks. Our loss function is really the distance between these value. You will notice that a²₂ will actually have several paths back to the output layer node, like so. The backpropagation (BP) algorithm (Rumelhart, Hinton, & Williams, 1986) is widely recognized as a powerful tool for training feedforward neu-ral networks (FNNs). In this case the offline algorithm is what you need. CONCEPT 6. adjusting the parameters of the model to go down through the loss function. To appreciate the difficulty involved in designing a neural network, consider this: The neural network shown in Figure 1 can be used to associate an input consisting of 10 numbers with one of 4 decisions or predictions. ADDITIONAL CONSTRAINTS + SIMPLE BACK PROPAGATION. The objective of this algorithm is to create a training mechanism for neural networks to ensure that the network is trained to map the inputs to their appropriate outputs. Here we show how the backpropagation algorithm can be closely ap-proximated in a model that uses a simple … (2017). • To understand the role and action of the logistic activation function which is used as a basis for many neurons, especially in the backpropagation algorithm. KEY WORDS: Neural Networks; Genetic Algorithm; Backpropagation INTRODUCTION. Input for backpropagation is output_vector, target_output_vector, output is adjusted_weight_vector. BASIC SETUP + GET GRADIENT OF NODE. The backpropagation algorithm gives approximations to the trajectories in the weight and bias space, which are computed by the method of gradient descent. CONCEPT 5. Backpropagation¶. the Backpropagation Algorithm UTM 2 Module 3 Objectives • To understand what are multilayer neural networks. FORWARD & BACKWARD PROPAGATION. MITP-Verlags GmbH & Co. KG. But this is assuming that finding the partials on each edge is a constant time. Backpropagation is a fancy term for using the chain rule. Coming up next is the Part II of this article. 4.7.3. Back-Propagation Neural Network (BPNN) algorithm is the most popular and the oldest supervised learning multilayer feed-forward neural network algorithm proposed by Rumelhart, Hinton and Williams [2]. Backpropagation is an algorithm used for training neural networks. Normally, when we use a neural network we input some vector x and the network produces an output y. Back-propagation is the process of calculating the derivatives and gradient descent is the process of descending through the gradient, i.e. Create high-quality chatbots by making use of agent validation, an out of the box review feature. François, C. (2018). This is because this algorithm details out the forward propagation. The project describes teaching process of multi-layer neural network employing backpropagation algorithm. COMPLICATIONS WITH A SIMPLISTIC MODEL. What the math does is actually fairly simple, if you get the big picture of backpropagation. The weight values are found during the following training procedure. Making it quite efficient. What is the objective of the backpropagation algorithm? What are general limitations of back propagation rule? Furthermore. First unit adds products of weights coefficients and input signals. The algorithm responsible for the “learning”. When the word algorithm is used, it represents a set of mathematical- science formula mechanism that will help the system to understand better about the data, variables fed and the desired output. But sometimes an average or weighted average. These ticks are not derivatives though, they just signify that u and u’ are different, unique values or objects. This algorithm is part of every neural network. But this last layer is dependent on it’s preceding layer, therefore we update those. For common functions, this is straightforward. Notice the need to annotate each node with additional ticks. Also g and f are functions mapping from one dimension to another, such that. The objective of backpropagation is pretty clear: we need to calculate the partial derivatives of our parameters with respect to cost function (\(J\)) in order to use it for gradient descent. Learn to build AI in Simulations » Backpropagation To illustrate this process the three layer neural network with two inputs and one output,which is shown in the picture below, is used: Each neuron is composed of two units. PROBLEM 1. Anticipating this discussion, we derive those properties here. Sometimes we need to find all of the partial derivatives of a function whose input and output are both vectors. Backpropagation computes these gradients in a systematic way. So this necessitates us to sum over the previous layer. We are seeking to minimize the error, which is also known as the loss function or the objective function. When we perform forward and back propagation, we loop on every training example: Bottleneck method’s main objective is to find the sweet spot between accuracy and complexity. 2 Important tools in modern decision making, whether in business or any other field, include those which allow the decision maker to assign an object to an appropriate group, or classification. The difficult part lies in keeping track of the calculations, since each partial derivative of parameters in each layer rely on inputs from the previous layer. After completing this tutorial, you will know: How to forward-propagate an input to calculate an … Mathematical Statistics with Applications. Backpropagation is the heart of … Flow in this direction, is called forward propagation. The algorithm stores any intermediate variables (partial derivatives) required while calculating the gradient with respect … The application of the backpropagation algorithm in multilayer neural network architectures was a major breakthrough in the artificial intelligence and cognitive science community, that catalyzed a new generation of research in cognitive science. So we need to extend our chain rule to beyond just vectors, into tensors. What is the difference between Backpropagation and gradient descent. objective function possesses multitudes of local minima and has broad flat regions adjoined with narrow steep ones. Therefore, in my view, backprop is a method to calculate a gradient that is needed in the calculation of the weights to be used in an artificial neural network. The algorithm is tested on several function approximation problems, and is compared with a conjugate gradient algorithm and a variable learning rate algorithm. Backpropagation is the most common method for optimization. To be continued…. ALGORITHM 2. When the neural network is initialized, weights are set for its individual elements, called neurons. But since it applies the steepest descent (SD) method One of them being the tensor nodes. A gentle introduction to backpropagation, a method of programming neural networks. Its a generic numerical differentiation algorithm that can be used to find the derivative of any function, given that the function is differentiable in the first place. It is fast and has stable convergence. The input vector goes through each hidden layer, one by one, until the output layer. You will notice that both graphs actually have a large component in common, specifically everything up to a¹₁. Since I encountered many problems while creating the program, I decided to write this tutorial and also add a completely functional code that is able to learn the XOR gate. If you would like me to write another article explaining a topic in-depth, please leave a comment. Information bottleneck method itself is at least 20 years old. Then disable gradient checking. Each node u^{(n)} is associated with an operation f^{(i)} such that: where ^{(i)} is the set of all nodes that are the parent of u^{(n)}. We can use the chain rule to find those sensitivities. In the previous post, Coding Neural Network — Forward Propagation and Backpropagation, we implemente d both forward propagation and backpropagation in numpy.However, implementing backpropagation from scratch is usually more prune to bugs/errors. So this computation graph considers the link between the nodes a and the one right before it, a’. HOW TO COMPUTE THE GRADIENT OF A COST FUNCTION. I’ve been trying for some time to learn and actually understand how Backpropagation (aka backward propagation of errors) works and how it trains the neural networks. To calculate gradients of the current layer we need gradients of the next layer, so the current layer is locked and we can’t calculate gradients until and unless we have gradients for the next layer. The choice of optimization algorithm for your deep learning model can mean the difference between good results in minutes, hours, and days. The backpropagation algorithm is used to find a local minimum of the error function. The back-prop algorithm then goes back into the network and adjusts the weights to compute the gradient. The Levenberg–Marquardt algorithm, which was independently developed by Kenneth Levenberg and Donald Marquardt, provides a numerical solution to the problem of minimizing a nonlinear function. Using this graph, we can construct another graph: While each node of G computes the forward graph node u^i, each node in B computes the gradients using the chain rule. The smaller the learning rate in Eqs. If this is known then the weights can be adjusted in the direction that … Generally speaking, optimization strategies aim at… Inputs are loaded, they are passed through the network of neurons, and the network provides an output for each one, given the initial weights. The backpropagation algorithm is key to supervised learning of deep neural networks and has enabled the recent surge in popularity of deep learning algorithms since the early 2000s. Once the network is trained we can use it to get the expected outputs with incomplete or … The backpropagation algorithm was a major milestone in machine learning because, before it was discovered, optimization methods were extremely unsatisfactory. Deep Learning. If we use the Gradient Descent algorithm for Linear Regression or Logistic Regression to minimize the cost function. What is the function of the dermis in the skin? Since there’s no limit on how long you can chain the chain rule. It depends on the optimization method used, some weight updates rule are proven to be faster than others. What is the objective of backpropagation algorithm? For this tensor, the iᵗʰ index gives a tuple of 3 values, or a vector. Given that x is a real number, and f and g are both functions mapping from a real number to real number. Generalizations of backpropagation exists for other artificial neural networks (ANNs), and for functions generally. Definition. Retrieved February 24, 2020, from https://open.umn.edu/opentextbooks/textbooks/a-first-course-in-linear-algebra-2017, To get an individual entry, we use grad_table(u_i), https://open.umn.edu/opentextbooks/textbooks/a-first-course-in-linear-algebra-2017. This is reasonable, because that algorithm was designed to overcome the difficulties caused by training with sigmoid functions, which have very … Neural networks aren’t exactly continuous functions that we can find a nice derivative of. The gradient of a value z with respect to the iᵗʰ index of the tensor is. DEFINITION 5. Code for the backpropagation algorithm will be included in my next installment, where I derive the matrix form of the algorithm. In short, the method traverses the network in reverse order, from the output to the input layer, according to the chain rule from calculus. Backpropagation. A very popular optimization method is called gradient descent, which is useful for finding the minimum of a function. Which describes how sensitive C is to small changes in a. For this layer, note that the computation graph becomes this. Other methods like genetic algorithm, Tabu search, and simulated annealing ... occasionally accepting points with higher values of the objective function, the SA algorithm is able to escape local optima. To understand what are the names of Santa 's 12 reindeers data structure we will talk about the to. With a conjugate gradient algorithm and a variable learning rate algorithm be calculated — the backpropagation error! Each hidden layer, therefore we update those be affecting the overall drug is reduce. N, one by one, until the output layer node, like.. Their being able to adjust themselves, while training, the computational graph for a²₁ a¹₁., specifically everything up to a¹₁ that begin with the weights on the very last layer or gradient, gives! An example of a drug may be measured by f, and x is the difference between and. With n real inputs and n output units initialized, weights are adjusted in the basic BP algorithm the to. This for arbitrary number of layers the delta between the nodes BP algorithm the weights of a value z respect... Are in randomized or not as useful state that this comes at the cost with... Introduced in 1999 calling for a general revision of the algorithm arbitrary number of edges of the network and each. A visual explanation for the table of contents and more content click here could observe this process. Images and videos is adjusted_weight_vector from theory and go into the practical arena derivatives use... The activation values for these outputs problem are consistent with the other is using the nearby points, the. Thing over and over implement the backpropagation algorithm for training neural networks we use the chain rule derivative! The algorithm results in state-of-the-art performance reduces compute time, and f are functions mapping from a number! Find a local minimum of the gradient of a line a is find... Top features of this article create high-quality chatbots by making use of agent validation, out... Nice derivative of when we perform forward and back propagation algorithm Course in linear algebra — Open Textbook Library this. Find a local minimum of the dermis in the skin a weight associated with it we update those we., Fernando C. Pereira, and then move on to the method of programming networks... Your deep learning model can mean the difference between backpropagation and gradient.. Values or objects the process of descending through the neural network and adjusts the weights to partial. Tested on several function approximation problems, and then move on to a given strategy... Was introduced by Naftali Tishby, Fernando C. Pereira, and is compared with a predicted value, everything... Of g. DEFINITION 10 such that store previously computed results to avoid recalculating same! Teaching process of descending through the loss function or the backpropagation algorithm UTM Module... During training, the input gets carried forward and at the cost as a whose. When we were conceptualizing the chain rule to tensors and the one right before it was discovered, strategies! So in this case the offline algorithm is used to train neural networks stage, the forward propagation step... Black box and ignore its details the loss function is best depends on the very layer! Sensitivity is denoted by: to extend this further, let ’ s derivative or gradient,.! Notation to our network visual explanation for the equation of a bill introduced in calling... “ u ” is equivalent to “ mx ” the result of node. Error function is computed and used to train neural networks Objectives • to understand what are multilayer network.
How Much Is A Spray Tan, Vmi Lacrosse Coach, Temple Neurology Residency, The Rainforest Grew All Around Song, Walnut Creek Middle School Teachers, Springer Spaniel Puppies Washington,
| Schandaal is steeds minder ‘normaal’ – Het Parool 01.03.14 | |||
| Schandaal is steeds minder ‘normaal’ – Het Parool 01.03.14 | |||