In a general artificial intelligence or machine learning, other than many computer based ad-hoc techniques, why are the exact or direct methods/algorithms not sufficiently good/efficient in yielding high accuracy, in comparison/contrast to the nonlinear approximate methods/algorithms?
It depends on application where you are using AI? Sometimes fuzzy algorithms also applicable for the applications. So may supervised and unsupervised algorithms are here. These algorithms are suitable for different data-sets.
If you mention your data-set and application then more precisely it ca be discuss. Thank you. Muhammad Ali
The difficulty arises mainly in computer vision tasks for high dimensional data (images or video clips). In such cases linear models are never efficient for high-dimensional data as well as large databases.
As Muhammad Ali said high dimensional data is the problem. This curse of dimensionality may not only arise due to a high input size (high-resolution images) but also in complex data or complex correlations like in LiDar data or NLP.
Here, in data analysis from high dimension data we mean too many variables (lights, weather, emotions, makeup, make over...... in case of images and video data) which causes effect in recognition of the given data.... Such high dimensional data is not preferred to be analyzed via linear models or short-cuts ad-hoc or heuristic methods. ......
The real-world data contain issues such as noise, high dimensionality, nonlinearity, imbalance, etc. These unwanted problematic complexities require proper preprocessing of data for its elimination before fitting any ML models.
- Badges
- Science topic