Searcharxiv⌕ Search

arXiv subjects

Takuya Akiba

Publications and source records attributed to Takuya Akiba.

29 records · Page 2Linked to original sources
arXiv2019

A Graph Theoretic Framework of Recomputation Algorithms for Memory-Efficient Backpropagation

Recomputation algorithms collectively refer to a family of methods that aims to reduce the memory consumption of the backpropagation by selectively discarding the intermediate results of the forward propagation and recomputing the discarded results as needed. In this paper, we will propose a novel and efficient recomputation method that can be applied to a wider range of neural nets than previous methods. We use the language of graph theory to formalize the general recomputation problem of minimizing the computational overhead under a fixed memory budget constraint, and provide a dynamic programming solution to the problem. Our method can reduce the peak memory consumption on various benchmark networks by 36%~81%, which outperforms the reduction achieved by other methods.

cs.LG↗
arXiv2019

Sampling Techniques for Large-Scale Object Detection from Sparsely Annotated Objects

Efficient and reliable methods for training of object detectors are in higher demand than ever, and more and more data relevant to the field is becoming available. However, large datasets like Open Images Dataset v4 (OID) are sparsely annotated, and some measure must be taken in order to ensure the training of a reliable detector. In order to take the incompleteness of these datasets into account, one possibility is to use pretrained models to detect the presence of the unverified objects. However, the performance of such a strategy depends largely on the power of the pretrained model. In this study, we propose part-aware sampling, a method that uses human intuition for the hierarchical relation between objects. In terse terms, our method works by making assumptions like "a bounding box for a car should contain a bounding box for a tire". We demonstrate the power of our method on OID and compare the performance against a method based on a pretrained model. Our method also won the first and second place on the public and private test sets of the Google AI Open Images Competition 2018.

cs.CV↗
arXiv2018

Adversarial Attacks and Defences Competition

To accelerate research on adversarial examples and robustness of machine learning classifiers, Google Brain organized a NIPS 2017 competition that encouraged researchers to develop new methods to generate adversarial examples as well as to develop new ways to defend against them. In this chapter, we describe the structure and organization of the competition and the solutions developed by several of the top-placing teams.

cs.CV↗
arXiv2018

Variance-based Gradient Compression for Efficient Distributed Deep Learning

Due to the substantial computational cost, training state-of-the-art deep neural networks for large-scale datasets often requires distributed training using multiple computation workers. However, by nature, workers need to frequently communicate gradients, causing severe bottlenecks, especially on lower bandwidth connections. A few methods have been proposed to compress gradient for efficient communication, but they either suffer a low compression ratio or significantly harm the resulting model accuracy, particularly when applied to convolutional neural networks. To address these issues, we propose a method to reduce the communication overhead of distributed deep learning. Our key observation is that gradient updates can be delayed until an unambiguous (high amplitude, low variance) gradient has been calculated. We also present an efficient algorithm to compute the variance with negligible additional cost. We experimentally show that our method can achieve very high compression ratio while maintaining the result model accuracy. We also analyze the efficiency using computation and communication cost models and provide the evidence that this method enables distributed deep learning for many scenarios with commodity environments.

cs.LG↗
arXiv2017

Extremely Large Minibatch SGD: Training ResNet-50 on ImageNet in 15 Minutes

We demonstrate that training ResNet-50 on ImageNet for 90 epochs can be achieved in 15 minutes with 1024 Tesla P100 GPUs. This was made possible by using a large minibatch size of 32k. To maintain accuracy with this large minibatch size, we employed several techniques such as RMSprop warm-up, batch normalization without moving averages, and a slow-start learning rate schedule. This paper also describes the details of the hardware and software of the system used to achieve the above performance.

cs.DC↗
arXiv2017

ChainerMN: Scalable Distributed Deep Learning Framework

One of the keys for deep learning to have made a breakthrough in various fields was to utilize high computing powers centering around GPUs. Enabling the use of further computing abilities by distributed processing is essential not only to make the deep learning bigger and faster but also to tackle unsolved challenges. We present the design, implementation, and evaluation of ChainerMN, the distributed deep learning framework we have developed. We demonstrate that ChainerMN can scale the learning process of the ResNet-50 model to the ImageNet dataset up to 128 GPUs with the parallel efficiency of 90%.

cs.DC↗
arXiv2016

Cut Tree Construction from Massive Graphs

The construction of cut trees (also known as Gomory-Hu trees) for a given graph enables the minimum-cut size of the original graph to be obtained for any pair of vertices. Cut trees are a powerful back-end for graph management and mining, as they support various procedures related to the minimum cut, maximum flow, and connectivity. However, the crucial drawback with cut trees is the computational cost of their construction. In theory, a cut tree is built by applying a maximum flow algorithm for $n$ times, where $n$ is the number of vertices. Therefore, naive implementations of this approach result in cubic time complexity, which is obviously too slow for today's large-scale graphs. To address this issue, in the present study, we propose a new cut-tree construction algorithm tailored to real-world networks. Using a series of experiments, we demonstrate that the proposed algorithm is several orders of magnitude faster than previous algorithms and it can construct cut trees for billion-scale graphs.

cs.DS↗
arXiv2016

Fractality of Massive Graphs: Scalable Analysis with Sketch-Based Box-Covering Algorithm

Analysis and modeling of networked objects are fundamental pieces of modern data mining. Most real-world networks, from biological to social ones, are known to have common structural properties. These properties allow us to model the growth processes of networks and to develop useful algorithms. One remarkable example is the fractality of networks, which suggests the self-similar organization of global network structure. To determine the fractality of a network, we need to solve the so-called box-covering problem, where preceding algorithms are not feasible for large-scale networks. The lack of an efficient algorithm prevents us from investigating the fractal nature of large-scale networks. To overcome this issue, we propose a new box-covering algorithm based on recently emerging sketching techniques. We theoretically show that it works in near-linear time with a guarantee of solution accuracy. In experiments, we have confirmed that the algorithm enables us to study the fractality of million-scale networks for the first time. We have observed that its outputs are sufficiently accurate and that its time and space requirements are orders of magnitude smaller than those of previous algorithms.

cs.DS↗
arXiv2014

Branch-and-Reduce Exponential/FPT Algorithms in Practice: A Case Study of Vertex Cover

We investigate the gap between theory and practice for exact branching algorithms. In theory, branch-and-reduce algorithms currently have the best time complexity for numerous important problems. On the other hand, in practice, state-of-the-art methods are based on different approaches, and the empirical efficiency of such theoretical algorithms have seldom been investigated probably because they are seemingly inefficient because of the plethora of complex reduction rules. In this paper, we design a branch-and-reduce algorithm for the vertex cover problem using the techniques developed for theoretical algorithms and compare its practical performance with other state-of-the-art empirical methods. The results indicate that branch-and-reduce algorithms are actually quite practical and competitive with other state-of-the-art approaches for several kinds of instances, thus showing the practical impact of theoretical research on branching algorithms.

cs.DS↗
arXiv2013

Fast Exact Shortest-Path Distance Queries on Large Networks by Pruned Landmark Labeling

We propose a new exact method for shortest-path distance queries on large-scale networks. Our method precomputes distance labels for vertices by performing a breadth-first search from every vertex. Seemingly too obvious and too inefficient at first glance, the key ingredient introduced here is pruning during breadth-first searches. While we can still answer the correct distance for any pair of vertices from the labels, it surprisingly reduces the search space and sizes of labels. Moreover, we show that we can perform 32 or 64 breadth-first searches simultaneously exploiting bitwise operations. We experimentally demonstrate that the combination of these two techniques is efficient and robust on various kinds of large-scale real-world networks. In particular, our method can handle social networks and web graphs with hundreds of millions of edges, which are two orders of magnitude larger than the limits of previous exact methods, with comparable query time to those of previous methods.

cs.DS↗