Ingest a Visual Research Article into the Wiki

Problem/Feature Description

A data science team maintains a shared knowledge base wiki for their research. A new team member, Priya, has written a technical summary on neural network architectures that includes several diagrams stored as local image files alongside the article. The team uses the wiki to compound knowledge across projects, so images referenced in wiki pages need to be stored locally under the wiki's assets folder—not referenced as external or raw-path links—so the wiki remains self-contained and portable.

Priya's article is ready in the raw sources folder. The team wants it ingested into the wiki, with all referenced images available under the wiki's asset folder (wiki/assets/) and wiki pages referencing them with relative paths. The existing wiki already has a few topic pages on related subjects that should be cross-linked where appropriate.

Output Specification

Ingest the raw source into the wiki. The expected outputs are:

• One or more new or updated wiki pages under wiki/topics/ covering the main concepts in the article (convolutional networks, residual connections)
• Images referenced in the article must be available within the wiki tree so wiki pages can reference them portably
• Updated wiki/index.md with links and one-line descriptions for new or changed pages
• A new entry appended to wiki/log.md using the correct heading format

Input Files

The following files are provided as inputs. Extract them before beginning.

=============== FILE: wiki/index.md ===============

Wiki Index

Topics

• Gradient Descent — Overview of gradient descent optimization and learning rate selection
• Backpropagation — How gradients flow through neural networks during training

Sources

• raw/2026-01-10-gradient-descent-overview.md

=============== FILE: wiki/topics/gradient-descent.md ===============

Gradient Descent

Gradient descent is an optimization algorithm used to minimize a loss function by iteratively moving in the direction of the steepest descent.

Learning rate

A learning rate of 0.01 is a common default starting point.

See also

• Backpropagation

=============== FILE: wiki/topics/backpropagation.md ===============

Backpropagation

Backpropagation is the algorithm used to compute gradients of a loss function with respect to model parameters using the chain rule.

It is used in conjunction with gradient descent to train neural networks.

=============== FILE: wiki/log.md ===============

[2026-01-10] ingest | gradient-descent-overview

• Created: wiki/topics/gradient-descent.md
• Created: wiki/topics/backpropagation.md
• Updated: wiki/index.md

=============== FILE: raw/2026-04-06-cnn-architectures.md ===============

Convolutional Neural Network Architectures

Date: 2026-04-06 Author: Priya Sharma

Overview

Convolutional Neural Networks (CNNs) use spatial feature extraction through learned filters. The core operation is the convolution, which slides a filter across the input to produce a feature map.

Key diagram showing the CNN pipeline:

Residual Connections

ResNets (He et al., 2016) introduced skip connections that allow gradients to flow directly across layers:

y = F(x) + x

This solved the vanishing gradient problem for very deep networks.

Residual block diagram:

Relationship to gradient descent

CNNs are trained with gradient descent and backpropagation. The depth enabled by residual connections makes the optimization landscape more well-behaved.

Key researchers

• Yann LeCun — pioneered CNNs for handwritten digit recognition (LeNet, 1989)
• Kaiming He — lead author of ResNet paper (2016)

=============== FILE: images/cnn-pipeline.png =============== <binary image file — treat this as an existing image file at path images/cnn-pipeline.png>

=============== FILE: images/residual-block.png =============== <binary image file — treat this as an existing image file at path images/residual-block.png>