Top 10 problems of Machine Learning in 2017

Machine Learning is a kind of Artificial Intelligence(AI) that helps

computers to learn without being programmed explicitly. It focuses

on development of computer programs that can change when

exposed to new data.

Machine Learning tasks are classified into 3 categories, depending on

the nature of learning signal or feedback of a learning system.Below is

the in depth explanation about it:

Supervised Learning: Here the teacher gives example inputs and their

desired outputs to the computer. The goal is to learn a rule that maps

inputs to outputs.

Unsupervised Learning: In this case nothing is given to the

computer; the learning system itself has to find structure in its input.

The goal is discover hidden patterns in the data.

Reinforcement Learning: Computer program must perform a certain

goal dynamically. The program gets feedback as rewards and

punishments.1.Natural Language processing: Understanding language is still a

challenge for even the deepest networks.

2. Differential Neural Computers: These are a special type of memory

augmented neural networks, which can think but cannot scale.

3.Memory augmented neural networks are a type of neural networks

which has a memory blocks which can be read and written to by the

network. We need to find the better way to discover facts, store and

use them effectively to solve problems.

4.Object Detection: Machine Learning cannot understand or detect

images. Object Recognition is still a problem for learning systems.

Many approaches have been implemented over decades.

5.Attention: Systems cannot grab attention in neural networks. So we

need to build attention mechanisms in neural networks to make them

better.

6.Machine learning cannot be learnt by observations and listening.

Computer should listen what the person is speaking and respond

accordingly.

7.One-shot learning: The ability of a memory augmented neural

network to rapidly receive and understand new data and use data to

make accurate predictions after few examples.

8.Effective Response Generation: The ability to generate contextual

responses. Computer should have a library of templates responses or

it should respond exactly about the particular enquirer.

9.Automated learning from a repository of resources: Learning from

other resources by making a graph connected sense is missing. In this

case, more automated intelligent system is required.

10.Facial Identification over varying feature space: Facial recognition

is not perfect over varying feature space even though it is a primary

requirement.