Input
In deep learning, input is the raw data you provide to the model so that it can learn or make predictions.
Inputs are usually represented as numerical values, but they can come from various sources like images, text, numbers, audio, or sensor data.
Detailed Explanation:
A neural network (deep learning model) is like a human brain that takes inputs and makes decisions.
- For humans, inputs come from eyes, ears, skin, etc.
- For machines, inputs are numbers fed into the model.
In deep learning:
- Inputs are organized into arrays or tensors (like tables of numbers).
- Each input can have one or multiple features (characteristics of the data).
- The model processes these inputs to learn patterns, relationships, or make decisions.
Mathematical View:
\[\large \text{Input} = [x_1, x_2, x_3, ..., x_n]\]Where:
- $x_1, x_2, x_3, …, x_n$ are input features.
Types of Inputs:
| Type | Example |
|---|---|
| Numeric | Age, salary, temperature |
| Text | Sentences, words, documents |
| Image | Pixels (converted to numbers) |
| Audio | Sound waves converted to numerical values |
| Video | Series of image frames + audio |
Real-Life Examples:
| Real-Life Scenario | Input for Deep Learning Model |
|---|---|
| Predicting house price | Size of the house, number of rooms, location, year built |
| Recognizing handwritten digits | Pixels of the image of the digit (0-9) |
| Language translation | Sentence in English (converted to numbers via embeddings) |
| Voice assistant (like Alexa) | Audio of spoken words (converted to numerical features) |
| Predicting if a customer will buy | Age, income, browsing history, past purchases |
Example:
Example 1: Cooking Recipe
- Input: Ingredients like flour, sugar, milk, etc.
- Model: Cook (processes the ingredients)
- Output: Cake or Bread
Example 2: ATM Machine
- Input: You enter your PIN and amount requested.
- Model: ATM system (processes your request)
- Output: Cash or transaction declined.
Example 3: Weather Prediction
- Input: Past temperature, humidity, wind speed.
- Model: Weather forecasting system.
- Output: Tomorrow’s weather.
Key Points:
- Input is what you give to the model.
- It can be simple (like a number) or complex (like an image or text).
- The quality and format of the input directly affect the model’s performance.
- Deep learning models expect inputs as numerical arrays, even if the original input is text, image, or audio.
import numpy as np
# Step 1: Define the input features (as an array)
# Example: Predicting house price using [size, number_of_rooms, location_score]
inputs = np.array([2000, 3, 8]) # e.g., 2000 sq ft, 3 rooms, location score 8
# Step 2: Define the weights (importance of each input)
weights = np.array([0.5, 0.3, 0.2]) # these are randomly assigned for demonstration
# Step 3: Define bias
bias = 5 # extra adjustment added to the output
# Step 4: Calculate the weighted sum
weighted_sum = np.dot(inputs, weights) + bias
# Step 5: Apply activation function (ReLU used here as example)
def relu(x):
return max(0, x)
output = relu(weighted_sum)
# Display the result
print(f"Input Features: {inputs}")
print(f"Weights: {weights}")
print(f"Weighted Sum (before activation): {weighted_sum}")
print(f"Output (after ReLU activation): {output}")
Input Features: [2000 3 8]
Weights: [0.5 0.3 0.2]
Weighted Sum (before activation): 1007.5
Output (after ReLU activation): 1007.5