Retraining Conditions: How Much to Retrain a Network After Pruning?
17 Dec 2019-pp 150-160
TL;DR: This paper analyses the fine-tuning or retraining step after pruning the network layer-wise and derives lower bounds for the number of epochs the network will take based on the amount of pruning done, and proposes a new parameter named ‘Net Deviation’ which can be used to estimate how good a pruning algorithm is.
Abstract: Restoring the desired performance of a pruned model requires a fine-tuning step, which lets the network relearn using the training data, except that the parameters are initialised to the pruned parameters. This relearning procedure is a key component in deciding the time taken in building a hardware-friendly architecture. This paper analyses the fine-tuning or retraining step after pruning the network layer-wise and derives lower bounds for the number of epochs the network will take based on the amount of pruning done. Analyses on the propagation of errors through the layers while pruning layer-wise is also performed and a new parameter named ‘Net Deviation’ is proposed which can be used to estimate how good a pruning algorithm is. This parameter could be an alternative to ‘test accuracy’ that is normally used. Net Deviation can be calculated while pruning, using the same data that was used in the pruning procedure. Similar to the test accuracy degradation for different amounts of pruning, the net deviation curves help compare the pruning methods. As an example, a comparison between Random pruning, Weight magnitude based pruning and Clustered pruning is performed on LeNet-300-100 and LeNet-5 architectures using Net Deviation. Results indicate clustered pruning to be a better option than random approach, for higher compression.
••01 Jan 1998
TL;DR: In this article, a graph transformer network (GTN) is proposed for handwritten character recognition, which can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters.
Abstract: Multilayer neural networks trained with the back-propagation algorithm constitute the best example of a successful gradient based learning technique. Given an appropriate network architecture, gradient-based learning algorithms can be used to synthesize a complex decision surface that can classify high-dimensional patterns, such as handwritten characters, with minimal preprocessing. This paper reviews various methods applied to handwritten character recognition and compares them on a standard handwritten digit recognition task. Convolutional neural networks, which are specifically designed to deal with the variability of 2D shapes, are shown to outperform all other techniques. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation recognition, and language modeling. A new learning paradigm, called graph transformer networks (GTN), allows such multimodule systems to be trained globally using gradient-based methods so as to minimize an overall performance measure. Two systems for online handwriting recognition are described. Experiments demonstrate the advantage of global training, and the flexibility of graph transformer networks. A graph transformer network for reading a bank cheque is also described. It uses convolutional neural network character recognizers combined with global training techniques to provide record accuracy on business and personal cheques. It is deployed commercially and reads several million cheques per day.
15 Feb 2016
TL;DR: Deep Compression as mentioned in this paper proposes a three-stage pipeline: pruning, quantization, and Huffman coding to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy.
Abstract: Neural networks are both computationally intensive and memory intensive, making them difficult to deploy on embedded systems with limited hardware resources. To address this limitation, we introduce "deep compression", a three stage pipeline: pruning, trained quantization and Huffman coding, that work together to reduce the storage requirement of neural networks by 35x to 49x without affecting their accuracy. Our method first prunes the network by learning only the important connections. Next, we quantize the weights to enforce weight sharing, finally, we apply Huffman coding. After the first two steps we retrain the network to fine tune the remaining connections and the quantized centroids. Pruning, reduces the number of connections by 9x to 13x; Quantization then reduces the number of bits that represent each connection from 32 to 5. On the ImageNet dataset, our method reduced the storage required by AlexNet by 35x, from 240MB to 6.9MB, without loss of accuracy. Our method reduced the size of VGG-16 by 49x from 552MB to 11.3MB, again with no loss of accuracy. This allows fitting the model into on-chip SRAM cache rather than off-chip DRAM memory. Our compression method also facilitates the use of complex neural networks in mobile applications where application size and download bandwidth are constrained. Benchmarked on CPU, GPU and mobile GPU, compressed network has 3x to 4x layerwise speedup and 3x to 7x better energy efficiency.
01 Jan 1989
TL;DR: A class of practical and nearly optimal schemes for adapting the size of a neural network by using second-derivative information to make a tradeoff between network complexity and training set error is derived.
Abstract: We have used information-theoretic ideas to derive a class of practical and nearly optimal schemes for adapting the size of a neural network. By removing unimportant weights from a network, several improvements can be expected: better generalization, fewer training examples required, and improved speed of learning and/or classification. The basic idea is to use second-derivative information to make a tradeoff between network complexity and training set error. Experiments confirm the usefulness of the methods on a real-world application.
••22 Jul 2015
TL;DR: In this article, the authors proposed a method to remove redundant neurons in a trained deep neural network (NN) model, which does not require access to any training/validation data.
Abstract: Deep Neural nets (NNs) with millions of parameters are at the heart of many state-of-the-art computer vision systems today. However, recent works have shown that much smaller models can achieve similar levels of performance. In this work, we address the problem of pruning parameters in a trained NN model. Instead of removing individual weights one at a time as done in previous works, we remove one neuron at a time. We show how similar neurons are redundant, and propose a systematic way to remove them. Unlike previous works, our pruning method does not require access to any training/validation data.
TL;DR: A new pruning method is developed, based on the idea of iteratively eliminating units and adjusting the remaining weights in such a way that the network performance does not worsen over the entire training set.
Abstract: The problem of determining the proper size of an artificial neural network is recognized to be crucial, especially for its practical implications in such important issues as learning and generalization. One popular approach for tackling this problem is commonly known as pruning and it consists of training a larger than necessary network and then removing unnecessary weights/nodes. In this paper, a new pruning method is developed, based on the idea of iteratively eliminating units and adjusting the remaining weights in such a way that the network performance does not worsen over the entire training set. The pruning problem is formulated in terms of solving a system of linear equations, and a very efficient conjugate gradient algorithm is used for solving it, in the least-squares sense. The algorithm also provides a simple criterion for choosing the units to be removed, which has proved to work well in practice. The results obtained over various test problems demonstrate the effectiveness of the proposed approach.
Related Papers (5)
14 Mar 2019