Machine Learning: Using the algorithm to parse the data, learn from it, and then make a decision and predict real-world events. That data used to train and learn how to complete the task from the data by the algorithm.
Machine learning techniques to achieve classification as shown below:
Traditional machine learning algorithms applied fingerprint identification, face detection, object detection field substantially reached the level of commercialization commercial requirements or particular scene. But the emergence of data representation and feature extraction problem (often required manual extraction characteristics ), proposed in order to further develop the depth of learning.
Deep learning can automatically learn the association between features and tasks can extract complex features from simple features. The initial depth study is the use of deep neural network (DNN) ( more than three a learning process of the neural network model) to resolve the feature expression. Depth is not a new concept of neural network itself, it may be generally understood as a neural network comprising a plurality of hidden layers. Process substantially as follows: initial weight values randomly , using the loss function as a basis for training cycle adjustment until find appropriate weights.
Traditional machine learning and deep learning of the main differences :
1, learning is not the same thing
Traditional machine learning to learn is a series set a good parameter index , indicators such as edge information, texture information, etc., to set these parameters into the system of weights training, return to optimal parameters. Therefore, the traditional machine learning is to learn some of the more specific parameters indicators, usually need to establish a model step by step to solve the problem. ( Splitting the problem into several parts, respectively, to solve them, and then the results combined together to get the answers you need)
深度学习是一种模糊的学习方式，机制是一个黑匣子，无法回答学习的一整个过程是在学习哪个指标，它学到的是一个整体特征。正如人脑神经网络有大量神经元复杂交错连接在一起传递电信号，深度学习的神经网络多层结构也是有很多节点来模仿神经元的功能传递信息，集中解决问题。
2、所需数据量和训练机制不同
传统机器学习适应各种数据量，尤其是数据量较小场景。当数据量到达某一程度时，传统机器学习的性能趋于稳定，此时输入再多数据，模型已经总体平稳。
若数据量迅速增加，深度学习的效果更为突出。对深度学习而言，送入数据量越多，系统就能学到更多需要的信息，更好理解信息的含义和内容并进行预测和判断。当然过，多数据量只会带来超长运算时间，越往后对性能提升微乎其微。
3、测试机制不同
传统机器学习需要将每个测试的数据进行和训练时类似的计算步骤，测试耗时长。相比较，深度学习测试时效率较高。
4、可解释性/可修改性不同
传统机器学习学到的是比较具体的信息一般有具体的物理含义，往往能够通过过程解释模型效果的好坏。可以根据特定的功能实现需求，针对性地调整模型的某个部分。
深度学习学到的是无法描述的数据信息，无法判断哪里出了问题使得结果变好还是变坏，大多根据经验调整模型。
（专设和毕设都和机器学习/深度学习有关(●′ω`●) 至于为什么这么有意识整理一下，这就要感谢我学校的毕设老师闹的小乌龙了ㄟ(◑‿◐ )ㄏ）