DNA家族:通过区块智能监督来促进重量共享NAS
IEEE transactions on pattern analysis and machine intelligence
|November 21, 2023
概括
蒸神经架构 (DNA) 技术通过解决大型搜索空间中不可靠的架构评级来提高重量共享神经架构搜索 (NAS). 这提高了自动机器学习的搜索效率和效率.
科学领域:
- 人工智能的人工智能
- 机器学习 机器学习
- 计算机视觉 计算机视觉
背景情况:
- 神经架构搜索 (NAS) 自动化了神经网络的设计,这是推进机器学习的关键方面.
- 权重共享NAS方法提供了效率,但由于在广的搜索空间内不可靠的架构评级,其搜索效率较低.
研究的目的:
- 通过开发更有效和可扩展的架构评级机制,克服重量共享NAS的局限性.
- 引入蒸神经架构 (DNA) 技术,以模块化搜索空间并提高NAS性能.
主要方法:
- 将大型搜索空间模块化成更小,易于管理的块.
- 使用蒸神经架构 (DNA) 技术开发了一系列模型.
- 利用了概括局限性工具来分析架构评级可信度.
主要成果:
- 在ImageNet上实现了移动卷积网络的78.9%,小型视觉变压器的83.6%的最先进的top-1精度.
- 证明了评价所有架构候选人的能力,与以前的子空间启发式方法不同.
- 在计算约束下,成功地寻找了不同深度和宽度的架构.
结论:
- 拟议的DNA家族有效地解决了重量共享NAS中的可扩展性,效率和多模式兼容性问题.
- DNA技术为值得信赖的架构评级提供了强大的解决方案,大大提高了NAS的有效性.
- 这些发现为改善神经架构评级和NAS优化提供了宝贵的见解.
相关概念视频
DNA as a Genetic Template
6.8K
6.8K
DNA Microarrays
17.5K
Microarrays are high-throughput and relatively inexpensive assays that can be automated to analyze large quantities of data at a time. They are used in genome-wide studies to compare gene or protein expression under two varied conditions, such as healthy and diseased states. Microarrays consist of glass or silica slides on which probe molecules are covalently attached through surface functionalization. Most commonly, the slides are prepared through the chemisorption of silanes to silica...
17.5K
Improving Translational Accuracy
2.6K
2.6K
From DNA to Protein
18.5K
The flow of genetic information in cells from DNA to mRNA to protein is described by the central dogma, which states that genes specify the sequence of mRNAs, which in turn specify the sequence of amino acids making up all proteins. The decoding of one molecule to another is performed by specific proteins and RNAs. Because the information stored in DNA is so central to cellular function, it makes intuitive sense that the cell would make mRNA copies of this information for protein synthesis...
18.5K
DNA Isolation
39.3K
DNA isolation protocols can be fast and straightforward or complex and time-consuming depending on the type and quality of DNA required for further processing. For example, plasmid DNA extraction is a bit more complicated than genomic DNA extraction because of the need for an appropriate lysis method to separate plasmid DNA from gDNA during isolation. However, for specific applications, such as long-range DNA sequencing that require a good yield of high- quality DNA samples, we need to follow...
39.3K
Genomic DNA in Eukaryotes
47.0K
Eukaryotes have large genomes compared to prokaryotes. To fit their genomes into a cell, eukaryotic DNA is packaged extraordinarily tightly inside the nucleus. To achieve this, DNA is tightly wound around proteins called histones, which are packaged into nucleosomes that are joined by linker DNA and coil into chromatin fibers. Additional fibrous proteins further compact the chromatin, which is recognizable as chromosomes during certain phases of cell division.
47.0K


