Machine learning / artificial neural network · BEGINNER LESSON
Deep learning model
A deep learning model is a machine learning model built from an artificial neural network (ANN), which is a layered set of connected artificial neurons that learns patterns from data. It uses several hidden layers to transform inputs into outputs and can learn from raw data without explicit mechanistic rules.
What it is and what is measured
A deep learning model is a type of machine learning model built from an artificial neural network (ANN). An ANN is a computer-based network of artificial neurons arranged in layers: an input layer, one or more hidden layers, and an output layer. In practice, the model measures patterns in data by passing signals through these layers and adjusting connection weights during training. The key thing being learned is how inputs relate to outputs, not a hand-written biological rule. Deep learning is defined by having three or more hidden layers. [1]
How it works
During training, the model sees examples and changes the weights on its connections so that later outputs better match the training data. Because the layers are stacked, the model can build increasingly complex internal representations. This lets it capture nonlinear relationships, learn directly from raw data, and work with large, high-dimensional, or multimodal datasets such as text, images, and omics data. A feedforward deep learning model can approximate any continuous function with arbitrary accuracy in theory if its weights are adjusted appropriately. [0,1]
[1] pmc.ncbi.nlm.nih.gov · articles · PMC12965040Deep‐Learning‐ versus Hypothesis‐Driven Modeling in Model ... [2] Reference Guide on Artificial Intelligence - NCBI - NIH
What it is good for
These models are often used when the data are complex and explicit mechanistic assumptions are missing or incomplete. Compared with more hypothesis-driven approaches, deep learning can improve predictive performance and can find predictors that were not part of the original mechanistic thinking. In genomic and related settings, deep learning can also speed inference after training, especially when the task is to map complex inputs to predicted outputs. [0,2]
[1] pmc.ncbi.nlm.nih.gov · articles · PMC12965040Deep‐Learning‐ versus Hypothesis‐Driven Modeling in Model ... [3] pmc.ncbi.nlm.nih.gov › articles › PMC13393291A Systematic Review: Deep Learning for Analyzing Genomic Data ...
Limitations and cautions
Deep learning models can be hard to interpret because they often act like black boxes. Their predictions may not extrapolate well beyond the observed data, and they can overfit, meaning they become too tailored to the training examples and perform less well on new data. Their quality depends strongly on the amount, quality, and representativeness of the training data. In some domains, training can also be computationally expensive, and performance can drop when there is a gap between simulated training data and real data. [0,1,2]
[1] pmc.ncbi.nlm.nih.gov · articles · PMC12965040Deep‐Learning‐ versus Hypothesis‐Driven Modeling in Model ... [2] Reference Guide on Artificial Intelligence - NCBI - NIH [3] pmc.ncbi.nlm.nih.gov › articles › PMC13393291A Systematic Review: Deep Learning for Analyzing Genomic Data ...
SEE THE IDEA
How a deep learning model turns data into predictions
Raw examples such as images, text, sequences, or other measurements enter the network.
Several layers of artificial neurons transform signals and learn internal representations.
Connection strengths are adjusted so outputs better match the training examples.
The final layer produces a classification, score, or other predicted result.
Input data flows through multiple hidden layers; training changes weights so the output becomes a better match to the target pattern.
A deep learning model learns by passing data through many layers and adjusting connections, rather than by using a single hand-coded rule.
TRY IT YOURSELF
Test your understanding
Choose an answer, then check the explanation.
PUT IT TOGETHER
Match each concept to its meaning
Drag a label onto its description, or select a label and then select a description. This also works with a keyboard.
4 pairs to match
Sources and further reading
This general lesson was generated from public source material. It explains a concept; it does not interpret a private paper or provide medical advice.