Forward Propagation and Backward Propagation | Neural Networks | How to train Neural Networks

Main ideas / concepts

The subtitles appear to be heavily corrupted/mixed with unrelated phrases, so the technical content is only partially identifiable. However, the video title clearly indicates that it teaches how to train neural networks using:

1. Forward propagation
   • Computing outputs by passing input activations through the network layers.
2. Backward propagation
   • Using the output error to adjust the network’s parameters (weights), typically to reduce loss over training iterations.

Methods / steps mentioned (as far as can be inferred)

The subtitles contain fragments that look like they refer to training flow and parameter adjustment, but the exact instructional text is mostly garbled. A coherent likely intended structure is:

• Initialize model parameters
  • Start with initial values for weights (subtitles include fragmentary references to “initializing” and “cost”).
• Forward propagation
  • Feed the input through the network to compute an output (subtitles reference “generate output” / “output”).
• Compute loss / cost
  • Evaluate how wrong the output is using a cost function (subtitles include “cost” and “graph vs …” fragments).
• Backward propagation
  • Propagate the error backward through the network to compute gradients (subtitles include “backward … locations,” “will reduce weight,” and “propagation” wording).
• Update weights
  • Adjust weights using gradients to reduce the cost/loss (subtitles include “reduce weight”).
• Repeat
  • Iterate the above steps over training examples/epochs (subtitles include nonsensical but repetitive timing fragments like “every …”).

Other noticeable content

• A large portion of the subtitles consists of repetitive channel/subscribe prompts and unrelated text.
• There are also seemingly random terms and names, likely introduced by caption errors.

Speakers / sources featured (identified from subtitles)

No clear speakers are reliably identified. The only “sources” that appear as fragments are names/handles that may be unrelated or corrupted:

• Dhillon (appears as “Dhillon” in text)
• Navneet (appears in text)
• Anil Vij (appears in text)
• WWE (appears as a word/fragment, likely not a speaker)
• Govinda (appears as a word/fragment, likely not a speaker)

Overall: No definitive speaker attribution can be confirmed from the provided subtitle text due to heavy transcription noise.