Deep learning for cognitive computing

近年来，对人工智能的探索取得了长足的进步，这在很大程度上要归功于我们通过所谓的机器学习向计算机提供可操作知识的能力的突破：只要有足够的例子，计算机就可以学习执行任务，并正确地概括新的案例。这些突破源于一种名为深度学习的机器学习方法，它的灵感来自大脑，并基于人工神经网络的早期工作。最初的进展是在模式识别方面，特别是语音识别和计算机视觉。这些领域已经被深度学习彻底改变，无论是在学术界还是在信息技术行业。该项目旨在将神经网络从传统的模式识别领域扩展到所谓的认知计算领域，这是IBM的一个战略领域，包括自然语言理解，推理，机器翻译和对话等人工智能研究领域，但也利用了计算机视觉和语音识别的进展。

The quest for artificial intelligence has advanced substantially in recent years, thanks in good part to breakthroughs in our ability to provide actionable knowledge to computers through what is called machine learning: provided with enough examples, the computer can learn to perform a task, and generalize correctly to new cases. These breakthroughs arose from a machine learning approach called deep learning, inspired from brains and based on older work on artificial neural networks. Initial progress has been in pattern recognition, especially for speech recognition and computer vision. These areas have been revolutionized by deep learning, both in academia and in the information technology industry. This project aims at contributing to the expansion of neural networks outside of their traditional pattern recognition domain into what is called cognitive computing, a strategic area for IBM, one that encompasses artificial intelligence research domains such as natural language understanding, reasoning, machine translation and dialogue, but also takes advantage of progress in computer vision and speech recognition.