bsatyakishore/intelligent-hacking

Learning to use and the abilities of milkcocoa and enebular by developing this simple app which uses keras

Intelligent Hacking

Idea inspired from Stealing PINs via Mobile Sensors. This project is created to demonstrate the power of ML with IoT. Uhuru's Milkcocoa is used to developed this app. Runs using SimpleRNN on the server or edge side. Collects accelerometer's data on mobile using js through Milkcocoa's js api.

Running the code:

Running the server client code:

$ git clone https://github.com/BSatyaKishore/Intelligent-Hacking $ cd Intelligent-Hacking $ python -m SimpleHTTPServer

Running the ML component:

To be ran on the device on which computation needs to be done

$ cd ML/SDK $ python Main.py

Generating training data:

In ML/SDK/Main.py:22 (Line 22 of ML/SDK/Main.py), make

Prediction = RNNPredicition.Prediction(self.TrainingData) to Prediction = GenerateTrainingData.Prediction(self.TrainingData)

The training data will be generated in ML/SDK/TrainingData. All the accelerometer data, the clicks and everything will be saved in this file.

Training RNN Model:

$ cd ML/SDK $ python RNN-Keras.py

and the model is saved in ML/SDK/RNN_model file.

Running your models:

$ cd ML/SDK

Create a file with your code in python say MyModel.py and create a function like def Prediction(Data): in it.

Whenever there is a need of predicition, this function will be called on a new thread with the Data.

def Prediction(TrainingData):
	_Temp = []
	TrainingData.sort(key=lambda tup: tup[0])
	for i in TrainingData:
		a = fun(i[1][1:-1].split(','))
		_Temp.append(a)
		oldI = i[0]
		if 'clickedId' in str(i):
			break
	TrainingData = _Temp
	TrainingDataX = TrainingData[-6:-1]
	pred = model.predict(np.array([TrainingDataX]))
	pa = pred[0].tolist()
	a = pa.index(max(pa))
	return str(a/6)+'_'+str(a%6)

is sample RNN model code. Little parsing also needs to be done in the same function.

Connecting this with the system:

In ML/SDK/Main.py:22 (Line 22 of ML/SDK/Main.py), make

Prediction = RNNPredicition.Prediction(self.TrainingData) to Prediction = MyModel.Prediction(self.TrainingData)

Demo:

Now, there are two important links:

1. http://localhost:8000/Demo for Demo site.
2. http://localhost:8000/Client for client site. To be opened on mobile phone or device with accelerometer. Note: Insert IP and port of the server if accessing remotely.

Requirements:

1. keras on theano backend
2. python
3. h5py

Have Fun.