Exploring Unexplored Tensor Network Decompositions for Convolutional Neural Networks

نویسندگان

چکیده

畳み込みニューラルネットワーク(CNN)は画像など特定ドメインのデータの処理に対して高い性能を発揮することが知られている.しかしながら同時に計算能力も必要とするため,CNNの軽量化は深層学習コミュニティにおいて広く行われてきた.本稿ではテンソル分解を使ったCNNの軽量化に焦点を当てる.まずCNNのコンポーネントである畳み込み層の演算が,複数テンソル間の線形演算の表現方法であるテンソルネットワークによって記述できることを示す.次に畳み込み層の軽量化がテンソル分解によって特徴づけられること,またその分解方法もテンソルネットワークによって記述できることを示す.最後に可能な分解を探索することによって,予測精度と時間/空間複雑さのトレードオフを実験的に比較する.その結果,いくつかの非線形分解は既存の分解を凌駕することがわかった.なお,本原稿は著者らの論文1)を和訳しわかりやすく解説したものである.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Convolutional Rectifier Networks as Generalized Tensor Decompositions

Convolutional rectifier networks, i.e. convolutional neural networks with rectified linear activation and max or average pooling, are the cornerstone of modern deep learning. However, despite their wide use and success, our theoretical understanding of the expressive properties that drive these networks is partial at best. On other hand, we have a much firmer grasp of these issues in the world ...

متن کامل

Convolutional Rectifier Networks as Generalized Tensor Decompositions

In this paper we analyze the expressiveness of networks, i.e.the functions they can realize, through the notion of grid tensors. Recall from sec. 4 that given templates x . . .x ∈ R, the grid tensor of a score function hy : (R) → R realized by some network, is defined to be a tensor of order N and dimension M in each mode, denoted A(hy), and given by eq. 3. In particular, it is a tensor holding...

متن کامل

Tensor graph convolutional neural network

In this paper, we propose a novel tensor graph convolutional neural network (TGCNN) to conduct convolution on factorizable graphs, for which here two types of problems are focused, one is sequential dynamic graphs and the other is cross-attribute graphs. Especially, we propose a graph preserving layer to memorize salient nodes of those factorized subgraphs, i.e. cross graph convolution and grap...

متن کامل

Analysis and Design of Convolutional Networks via Hierarchical Tensor Decompositions

The driving force behind convolutional networks – the most successful deep learning architecture to date, is their expressive power. Despite its wide acceptance and vast empirical evidence, formal analyses supporting this belief are scarce. The primary notions for formally reasoning about expressiveness are efficiency and inductive bias. Expressive efficiency refers to the ability of a network ...

متن کامل

Boosting Dilated Convolutional Networks with Mixed Tensor Decompositions

Expressive efficiency is a concept that allows formally reasoning about the representational capacity of deep network architectures. A network architecture is expressively efficient with respect to an alternative architecture if the latter must grow super-linearly in order to represent functions realized by the former. A well-known example is the exponential expressive efficiency of depth, name...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

ژورنال

عنوان ژورنال: The Brain & Neural Networks

سال: 2022

ISSN: ['1340-766X', '1883-0455']

DOI: https://doi.org/10.3902/jnns.29.193