当前位置:网站首页>MIT博士论文 | 使用神经符号学习的鲁棒可靠智能系统
MIT博士论文 | 使用神经符号学习的鲁棒可靠智能系统
2022-07-06 00:33:00 【智源社区】

这篇论文表明,通过神经符号模型的视角来看待智能系统比传统的深度学习方法有几个好处。神经符号模型包含符号程序性构造,如循环、条件和连续的神经成分。符号部分使模型具有可解释性、泛化性和稳健性,而神经部分处理智能系统的复杂性。具体而言,本文提出了两类神经符号模型——状态机和神经符号transformers,并以基于强化学习的自主系统和多机器人系统为例对它们进行了评估。这些案例研究表明,学习的神经符号模型是人类可读的,可以外推到看不见的场景,并可以处理规范中的稳健目标。为了有效地学习这些神经符号模型,我们引入了利用机器学习和程序合成的最新技术的神经符号学习算法。
边栏推荐
- Hudi of data Lake (2): Hudi compilation
- MySql——CRUD
- MySQL global lock and table lock
- Pointer pointer array, array pointer
- [Chongqing Guangdong education] Chongqing Engineering Vocational and Technical College
- [designmode] Decorator Pattern
- Intranet Security Learning (V) -- domain horizontal: SPN & RDP & Cobalt strike
- 如何解决ecology9.0执行导入流程流程产生的问题
- STM32按键消抖——入门状态机思维
- 数据分析思维分析方法和业务知识——分析方法(三)
猜你喜欢

Browser reflow and redraw

Recognize the small experiment of extracting and displaying Mel spectrum (observe the difference between different y_axis and x_axis)
![[binary search tree] add, delete, modify and query function code implementation](/img/38/810a83575c56f17a7a0ed428a2e02e.png)
[binary search tree] add, delete, modify and query function code implementation
![Atcoder beginer contest 258 [competition record]](/img/e4/1d34410f79851a7a81dd8f4a0b54bf.gif)
Atcoder beginer contest 258 [competition record]

【DesignMode】装饰者模式(Decorator pattern)
![[noi simulation] Anaid's tree (Mobius inversion, exponential generating function, Ehrlich sieve, virtual tree)](/img/d6/c3128e26d7e629b7f128c551cd03a7.png)
[noi simulation] Anaid's tree (Mobius inversion, exponential generating function, Ehrlich sieve, virtual tree)

认识提取与显示梅尔谱图的小实验(观察不同y_axis和x_axis的区别)

OpenCV经典100题

Basic introduction and source code analysis of webrtc threads

如何利用Flutter框架开发运行小程序
随机推荐
The relationship between FPGA internal hardware structure and code
Solve the problem of reading Chinese garbled code in sqlserver connection database
2022-02-13 work record -- PHP parsing rich text
[designmode] Decorator Pattern
Global and Chinese markets for hinged watertight doors 2022-2028: Research Report on technology, participants, trends, market size and share
【EI会议分享】2022年第三届智能制造与自动化前沿国际会议(CFIMA 2022)
Gd32f4xx UIP protocol stack migration record
[QT] QT uses qjson to generate JSON files and save them
anconda下载+添加清华+tensorflow 安装+No module named ‘tensorflow‘+KernelRestarter: restart failed,内核重启失败
Ffmpeg learning - core module
Data analysis thinking analysis methods and business knowledge -- analysis methods (II)
Recognize the small experiment of extracting and displaying Mel spectrum (observe the difference between different y_axis and x_axis)
Hudi of data Lake (2): Hudi compilation
Wechat applet -- wxml template syntax (with notes)
Global and Chinese market of digital serial inverter 2022-2028: Research Report on technology, participants, trends, market size and share
Introduction of motor
MySql——CRUD
Natural language processing (NLP) - third party Library (Toolkit):allenlp [library for building various NLP models; based on pytorch]
LeetCode 1598. Folder operation log collector
QT -- thread