効率的なアテンションとフレッシェ距離による微分可能な敵対的生成ネットワークアーキテクチャ検索
Yandan Xu1, Qianjin Zhang1, Yu Xue2
1Huzhou Vocational and Technical College, 299 Xuefu Road, Huzhou 313000, P. R. China.
International journal of neural systems
|January 23, 2026
まとめ
EAMGANは、効率的なアテンションモジュールとフレッシェ・インセプション・ディスタンス(FID)ガイダンスを使用することで、敵対的生成ネットワーク(GAN)アーキテクチャ検索を改善します。この手法は、既存の手法と比較して、画質と検索効率を向上させます。
科学分野:
- 人工知能
- コンピュータビジョン
背景:
- 敵対的生成ネットワーク(GAN)アーキテクチャ検索は、高性能ネットワークの設計を自動化します。
- DAMGANのような既存の微分可能な検索手法は、非効率的なSENetモジュールとアテンショントレーニングのための敵対的損失により、検索効率とアーキテクチャ品質に影響を与える限界に直面しています。
研究 の 目的:
- EAMGAN、強化された微分可能なGANアーキテクチャ検索手法を提案すること。
- 効率的なアテンションとフレッシェ・インセプション・ディスタンス(FID)ガイダンスを組み込むことにより、検索効率とアーキテクチャ品質を向上させること。
主な方法:
- EAMGANを導入し、軽量なECA-Netアテンションモジュールを利用して、SENetの非効率的なコンポーネントを1D畳み込みで置き換え、パラメータを削減しました。
- アテンショントレーニングを敵対的最適化から分離し、アーキテクチャ選択を直接ガイドするために、カスタマイズされたFIDベースの損失関数を採用しました。
- CIFAR-10およびSTL-10データセットでパフォーマンスを評価しました。
主要な成果:
- EAMGANは、CIFAR-10でDAMGANを上回り、Inception Score(IS)を向上させ、FIDを低下させました。
- 検索コストの削減(DAMGANの0.09 GPU日に対して0.08 GPU日)で、優れた結果を達成しました。
- STL-10データセットで競争力のあるパフォーマンスと転移性を示しました。
結論:
- EAMGANは、GANアーキテクチャ検索に対して、より効率的で効果的なアプローチを提供します。
- 提案手法は、画像生成品質を向上させ、計算オーバーヘッドを削減します。
- EAMGANは、生成モデリングにおけるより広範なアプリケーションの可能性を示しています。
関連する概念動画
Distance Problem
52
When an object's velocity changes over time, the total distance traveled can be determined by summing small displacement intervals over short increments. This approach approximates the true distance through numerical summation and the use of integral calculus. An estimate of the total displacement can be obtained by measuring velocity at regular intervals and multiplying each value by the corresponding time step.If a runner accelerates over the first three seconds of a race, speed measurements...
52
Distance Corrections
285
To achieve precise distance measurements, especially in surveying and construction, certain corrections must be applied to account for potential sources of error like the standardization errors, temperature variations, and slope adjustments.Standardization error emerges when measurement equipment undergoes changes, such as wear, repairs, or weather impacts. To address this, surveyors compare the equipment’s readings to a standard. This process identifies any deviation that might lead to...
285
The Distance Formula
600
In geometry, measuring the direct distance between two points on a plane is essential in various practical and theoretical applications. Whether in navigation, engineering, or computer graphics, determining the shortest path between two locations involves using the distance formula. This formula is derived from the Pythagorean Theorem, which relates the lengths of the sides of a right triangle. On a coordinate plane, the horizontal and vertical distances between two points serve as the legs of...
600
Short-distance Transport of Resources
17.6K
Short-distance transport refers to transport that occurs over a distance of just 2-3 cells, crossing the plasma membrane in the process. Small uncharged molecules, such as oxygen, carbon dioxide, and water, can diffuse across the plasma membrane on their own. In contrast, ions and larger molecules require the assistance of transport proteins due to their charge or size. Transport across membranes also occurs within individual cells, playing a variety of essential roles for the plant as a whole.
17.6K
Distance Measurements by Taping
430
Tapes are essential in surveying for accurate, durable, and short-distance measurements. Made from lightweight, nylon-coated steel, they offer flexibility and strength for rugged outdoor use. The nylon coating protects against rust and wear, extending the tape's life. Standard lengths, around 30 meters, are marked in meters and millimeters for precision.Surveyors select tapes based on site conditions and accuracy needs. Lightweight, nylon-coated tapes are commonly used for ease of handling and...
430
Protein Networks
4.5K
An organism can have thousands of different proteins, and these proteins must cooperate to ensure the health of an organism. Proteins bind to other proteins and form complexes to carry out their functions. Many proteins interact with multiple other proteins creating a complex network of protein interactions.
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
These interactions can be represented through maps depicting protein-protein interaction networks, represented as nodes and edges. Nodes are circles that are representative of a protein,...
4.5K


