CAREER: Towards Better Understanding, Robustness, and Efficiency of Deep Learning

深度学习是当前许多智能决策系统背后的主要驱动力，并在图像处理、语音识别、语言翻译和游戏等各种应用中取得了前所未有的成功。然而，缺乏足够的理论理解限制了在现实环境中充分发挥深度学习潜力的能力，例如在对安全敏感或资源有限的情况下。该项目旨在提供一种全面和系统的方法，以了解在捕捉现实世界问题属性的数据模型下，实用的深度学习模型为什么会成功。该项目将开发这样一种方法，并利用揭示的原理来设计更健壮和更有效的深度学习方法。该项目将开发导致经验成功的实际数据的属性的新理论模型，并提供证明性能保证的框架，包括在存在对抗性攻击或有限标签数据的情况下的学习。它还将设计新的学习方法，这些方法被证明更健壮，标签数据效率更高。尽管最近开展了大量研究活动，但这一方向在很大程度上仍未得到探索。建议的理论和算法解决方案可以通过机器学习、统计和优化的跨学科混合工具来实现。拟议的计划是基于研究人员先前的工作，包括理论结果和经验验证。如果成功，这项拟议的研究将为未来的发展奠定基础，从而为现代智能系统带来变革。它还将有助于从实践中解决当前理论无法充分解决的新理论问题，并将对机器学习和优化产生持久影响。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Deep learning is a primary driving force behind many current intelligent decision-making systems and has achieved unprecedented success in various applications such as image processing, speech recognition, language translation, and game playing. However, the lack of adequate theoretical understanding is limiting the capacity to fully exploit the potential of deep learning in realistic environments, such as in security-sensitive or resource-constrained scenarios. This project aims to provide a thorough and systematic approach to understanding why practical deep-learning models succeed, under models of the data capturing the properties of real-world problems. This project will develop such an approach and employ the revealed principles in designing more robust and efficient deep-learning methods.The project will develop new theoretical models of properties of practical data leading to empirical success, and also provide frameworks for proving performance guarantees, including for learning in the presence of adversarial attacks or limited labeled data. It will also design new learning methods that are provably more robust and labeled-data efficient. This direction is still largely unexplored, despite significant recent research activities. The proposed theoretical and algorithmic solutions are possible through an interdisciplinary mix of tools from machine learning, statistics, and optimization. The proposed program is grounded in the investigator's prior work that includes both theoretical results and empirical validation. If successful, the proposed research can be transformational for modern intelligent systems by laying the foundations for further development. It will also help to solve new theoretical problems from practice that are not adequately addressed by current theory and will have lasting impacts on machine learning and optimization.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.