AI combat 2019BUAA_ Xing Hang _ the eighth time job (s)

AI combat 2019BUAA_ Xing Hang _ the eighth time job (s)

---

　1. navigation

project	content
This work belongs courses	AI combat 2019
Where this requirement in the job	Here
My aim in this course is	Understand and can apply some common knowledge of artificial intelligence

2. Job Content

• How to identify illegal input
  to identify CNN at the time, would take logits softmax and argmax largest class, and if the character to determine accurately the value of other components should be small, but there is a particularly striking that the correct output, time when the illegal short legs can be output, the value of each component would be closer, then the model is difficult to determine what characters when input should therefore poor can be determined by calculating whether the components between the illegal input, if the electrode the difference is small, it is likely illegal input, if greater, the input should be legal.
• How to identify sin function
  for letter input occurs, there should be a unique data set is determined, and letters related to other functions should be placed in this dataset
• How to identify more complex expression structures (vertical structure)
  may be determined prior to the input character, adding a process to split the expression, to the expression judging portion, a lower portion is recognizing numeric characters, if the index can be considered the beginning or input, positioning or semicolon fractional index structure, recognition is completed. If the division does not recognize, it is identified in accordance with the normal input. In order to improve the speed, to avoid too frequent structure is divided, and upper and lower portions of the frequent identification of a common expression, can define a height threshold, when exceeded a certain height, divided Analyzing
• How to identify the structure surrounding the
  key should recognize such a structure that the operation symbol, such as a square root symbol, this symbol is determined whether there is a need, according to the results and then decide whether to identify the surrounding structure. If there exists surrounding the symbols and after successful identification, it may be considered after removed, the remaining part as a normal expression recognition.