Fall 2025 - UDel PHYS 461-66
dr. federica bianco 

@fedhere

fbianco@udel.edu

this slide deck:

https://slides.com/federicabianco/dsps_14

WRITTEN FINAL: 

72 hours at home

Dec 11 8AM - Dec 14 8AM

WORK ALONE

USE AI AT YOUR OWN RISK

USE THE SLACK CHANNEL FOR QUESTIONS 

DM ME IF YOUR QUESTION REQUIRES SHARING CODE OR SOLUTIONS

WRITTEN FINAL: 

72 hours at home

Dec 11 8AM - Dec 14 8AM

WORK ALONE

USE AI AT YOUR OWN RISK

USE THE SLACK CHANNEL FOR QUESTIONS 

DM ME IF YOUR QUESTION REQUIRES SHARING CODE OR SOLUTIONS

ORAL FINAL:

Schedule 30 minutes with me 

in those 30 minutes I will ask you questions about things you did in your final

You shall demonstrate that :
- you understand what you did
- you made informed choices based on what you leanred in the class

- you can answer questions about data science topics (e.g. what other method could you have used? or why would a NN not be a good choice here?)

https://calendly.com/fbbianco/30min

FINAL!!

Study Material:
conceptually

-> review slides, look at quiz questions you got wrong

Coding and models:

-> review old assignments and labs. Note that "solutions" to the HWs are posted int he HW folder on https://github.com/fedhere/DSPS_FBianco

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**
• assess performance of the model**
• compare models if more than one**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**
• assess performance of the model**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**
• assess performance of the model**
• compare models if more than one**

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**
• assess performance of the model**
• compare models if more than one**

all ** tasks need justification explanation and interpretation!

and all plots need  WHAT to count?

Tasks you will need to complete in the final:

• get your data - mind reproducibility!! e.g. get it from a stable existing link or put it on your google drive in a folder DSPS_final etc
• explore your data!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!**
• prepare your data for modeling**
• build your model including setting hyperparameters**
• train your model (if it is a ML model)**
• assess performance of the model**
• compare models if more than one**

all ** tasks need justification explanation and interpretation!

and all plots need  captions that described the "what" the "how" and the "WOW"

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

Higgs Boson

used -999 as missing value! ->

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

ATTENTION! exploratory data analysis is critical!!!

tasks include:

- look at datatypes
- ascertaining presence of NaN or missing values (e.g. -999) and devising and justifying a strategy to deal with them

- ascertaining the distribution of imput features

- ascertaining the distribution of target features! (e.g. howo many instances of each class, is the target value balanced??)
- ascertaining the correlation and covariance structure of the data!

And for all of this:

WHAT DOES IT MEAN?
HOW DOES IT GUIDE YOUR DATA PREP CHOICES?
HOW DO YOU USE IT TO CHOOSE YOUR MODEL?

HOW DOES IT INFLUENCE YOUR RESULT?

White Males: 43

White Female: 6

Non-White Males: 12

Non-White Female: 1

 Faces95 - Computer Vision Science Research Projects, Dr Libor Spacek.

The dataset was BIASED -> the model was BIASED

The model was not transferable

The central dogma of molecular biology is: Structure Determines Function.

• Enzymes: The precise 3D shape creates an "active site" where specific chemical reactions are catalyzed. The wrong shape means no reaction.

•  

• Antibodies: Their Y-shaped structure allows them to recognize and bind to foreign invaders like viruses and bacteria.

•  

• Structural Proteins (e.g., Collagen, Keratin): Their folded shapes provide strength and support to tissues like skin, hair, and bones.

•  

• Transport Proteins (e.g., Hemoglobin): Their shape allows them to pick up and release oxygen in the blood.

Proteins are:

Polymers of amino acids. The sequence of amino acids is recorded in your DNA

The amino acid sequence determines the folded structure of the protein

The folded structure has evolved to support the function of the protein

The information needed for folding is encoded in the protein's amino acid sequence, which itself is encoded in DNA.

1. 1 - Amino Acid Sequence (Primary Structure): The arrangement of amino acids in a protein. The DNA code is translated into a linear chain of amino acids. Proteins can be made from  20 different amino acids, and the structure and function of each protein are determined by the kinds of amino acids used to make it and how they are arranged.

2.  

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

send out groups of 100 proteins and ask researchers to predict folding

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

AlphaFOLD2

The information needed for folding is encoded in the protein's amino acid sequence, which itself is encoded in DNA.

1. 2 Local Folding (Secondary Structure): Sections of the chain spontaneously form local, stable patterns held together by hydrogen bonds. The most common are:

   • α-helices: A coiled, spring-like structure.

   • β-sheets: Pleated strands that line up side-by-side.

   •  

   • It is the way a polypeptide folds in a repeating arrangement.

   • This folding is a result of H bonding between the amine and carboxyl groups of non-adjacent amino acids

   •  

2.  

3.  

The information needed for folding is encoded in the protein's amino acid sequence, which itself is encoded in DNA.

1. 3 Global Folding (Tertiary Structure): The entire chain folds further into a unique, compact 3D shape.

2. It is caused by interactions between R groups: interactions between the amino acid side chains (hydrophobic effects, hydrogen bonds, ionic bonds, van der Waals forces).

3. •  

4.  

5.  

The information needed for folding is encoded in the protein's amino acid sequence, which itself is encoded in DNA.

1. 4 Complex Assembly (Quaternary Structure): Some proteins are made of multiple folded polypeptide chains (subunits) that assemble together to form the final, functional protein (e.g., hemoglobin).

2. •  

3.  

4.  

• Drug Design (Structure-Based Drug Design): Most drugs work by binding to a specific protein to either activate or block its function: design drugs to perfectly fit into a protein active site.

  • Example: Predicting the structure of the SARS-CoV-2 spike protein was crucial for rapidly developing vaccines and therapeutic antibodies

• Understanding Genetic Diseases

•  

• Fighting "Misfolding" Diseases: Alzheimer's and Parkinson's,

Motivation

Anfinsen’s dogma

In standard physiological environment, a protein’s structure is determined by the sequence of amino acids that make it up (1972 Nobel Prize in Chemistry).

If so we should be able to reliably predict a protein’s structure from its sequence.

Levinthal’s paradox

In the 1960s, Cyrus Levinthal showed that finding the native folded state of a protein by a random search among all possible configurations can take a time comparable with the lifetime of the Universe. 

35 aminoacids ->1e33 ways to fold

A small and physically reasonable energy bias against locally unfavorable configurations, of the order of a few kT, can reduce Levinthal's time to a biologically significant size.

For decades, predicting a protein's 3D structure from its amino acid sequence alone (the "protein folding problem") was one of the grand challenges in biology. This was revolutionized in 2020 by DeepMind's AlphaFold, an artificial intelligence system.

1600+ citations

2100+ citations

Levnithal's paradox

In the 1960s Cyrus Levinthal shiwed that there is a large number of possible conformation that a protein could theoretically adopt and if a protein were just to try them all it would take a time comparable to the life of the Universe (~10 Gyears)

ok, some are obviously better than others, so you dont have to explore them ALL

140,000 protein structures found by 2014

AlphaFold greatly improves the accuracy of structure prediction by incorporating novel neural network architectures and training procedures based on the evolutionary, physical and geometric constraints of protein structures.

new equivariant attention architecture, use of intermediate losses to achieve iterative refinement of predictions, masked MSA loss to jointly train with the structure, learning from unlabelled protein sequences using self-distillation and self-estimates of accuracy.

 In particular, we demonstrate a new architecture to jointly embed multiple sequence alignments (MSAs) and pairwise features, a new output representation and associated loss that enable accurate end-to-end structure prediction,

AlphaFold

  https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb

  https://colab.research.google.com/github/sokrypton/ColabFold/blob/main/AlphaFold2.ipynb

Innovative Input and Ouptut

Multi Sequence Allignment

Innovative Input and Ouptut

Multi Sequence Allignment

pairs preserved in multiple spicies: coevolution

Innovative Input and Ouptut

Multi Sequence Allignment

changes that occurr in pairs

pairs preserved in multiple spicies: coevolution

Innovative Input and Ouptut

Multi Sequence Allignment

Distance Matrix

changes that occurr in pairs

pairs preserved in multiple spicies: coevolution

G-A close together in the final structure

Innovative Input and Ouptut

Multi Sequence Allignment

Distance Matrix

changes that occurr in pairs

pairs preserved in multiple spicies: coevolution

G-A close together in the final structure

D-E are distant in the final structure

Innovative Input and Ouptut

Multi Sequence Allignment

Distance Matrix + Torsion Angles

changes that occurr in pairs

pairs preserved in multiple spicies: coevolution

The Distogram & 3D Coordinates

Instead of directly predicting 3D coordinates, AlphaFold's neural network was trained to predict two intermediate representations:

• Distogram: A matrix of pairwise distances between every amino acid residue in the protein. This defines the protein's fold topology.

• Torsion Angles: The internal rotation angles of the protein backbone (phi and psi angles).

CASP gives 1 sequence

additional input:

examples of similar

protein sequences

in their folded form

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

AlphaFold ->

2018

new architecture to jointly embed multiple sequence alignments (MSAs) and pairwise features

A 3D tensor of shape

(N_seq, N_res, c_m).

c_m =256 in AlphaFold2

CASP gives 1 sequence

Multiple Sequence Alignment

CASP gives 1 sequence

templates of folded proteins

A 3D tensor of shape

(N_res, N_res, c_z)

We show evidence in ‘Interpreting the neural network’ that a concrete structural hypothesis arises early within the Evoformer blocks and is continuously refined. The key innovations in the Evoformer block are new mechanisms to exchange information within the MSA and pair representations that enable direct reasoning about the spatial and evolutionary relationships.

Evoformer

Evoformer

Evoformer

Structure module

structure module that introduces an explicit 3D structure in the form of a rotation and translation for each residue of the protein (global rigid body frames).

Rotational - Translational - Physico-chemical constraints (think of a PINN!!)

Structure module

structure module that introduces an explicit 3D structure in the form of a rotation and translation for each residue of the protein (global rigid body frames).

Rotational - Translational - Physico-chemical constraints (think of a PINN!!)

AlphaFold3: diffusion NN

Generative AI dsps_09

new equivariant attention architecture, use of intermediate losses to achieve iterative refinement of predictions, masked MSA loss to jointly train with the structure, learning from unlabelled protein sequences using self-distillation and self-estimates of accuracy.

 In particular, we demonstrate a new architecture to jointly embed multiple sequence alignments (MSAs) and pairwise features, a new output representation and associated loss that enable accurate end-to-end structure prediction,

• How it works: It uses a deep learning network trained on the thousands of known protein structures in the Protein Data Bank (PDB). It looks for evolutionary patterns and physical constraints to predict the 3D coordinates of every atom.

•  

AlphaFOLD2 ->

2020

2018

Levnithal's paradox

In the 1960s Cyrus Levinthal shiwed that there is a large number of possible conformation that a protein could theoretically adopt and if a protein were just to try them all it would take a time comparable to the life of the Universe (~10 Gyears)

ok, some are obviously better than others, so you dont have to explore them ALL

140,000 protein structures found by 2014

Levnithal's paradox

In the 1960s Cyrus Levinthal shiwed that there is a large number of possible conformation that a protein could theoretically adopt and if a protein were just to try them all it would take a time comparable to the life of the Universe (~10 Gyears)

ok, some are obviously better than others, so you dont have to explore them ALL

140,000 protein structures found by 2014

200,000,000 protein structures found by AlphaFols2/3

https://www.youtube.com/watch?v=P_fHJIYENdI

models trained extensively on large amounts of data to solve generic problems - typically selfsupervised (i.e. trained as generative AI to reproduce input data, dsps_12)

Foundational AI models

We use the ILSVRC-2012 ImageNet dataset with 1k classes
and 1.3M images, its superset ImageNet-21k with
21k classes and 14M images and JFT with 18k classes and
303M high-resolution images.

Typically, we pre-train ViT on large datasets, and fine-tune to (smaller) downstream tasks.

For this, we remove the pre-trained prediction head and attach a zero-initialized D × K feedforward
layer, where K is the number of downstream classes

models trained extensively on large amounts of data to solve generic problems - typically selfsupervised (i.e. trained as generative AI to reproduce input data)

Foundational AI models

We use the ILSVRC-2012 ImageNet dataset with 1k classes
and 1.3M images, its superset ImageNet-21k with
21k classes and 14M images and JFT with 18k classes and
303M high-resolution images.

Typically, we pre-train ViT on large datasets, and fine-tune to (smaller) downstream tasks.

For this, we remove the pre-trained prediction head and attach a zero-initialized D × K feedforward
layer, where K is the number of downstream classes

models trained extensively on large amounts of data to solve generic problems - typically selfsupervised (i.e. trained as generative AI to reproduce input data)

Foundational AI models

We use the ILSVRC-2012 ImageNet dataset with 1k classes
and 1.3M images, its superset ImageNet-21k with
21k classes and 14M images and JFT with 18k classes and
303M high-resolution images.

Typically, we pre-train ViT on large datasets, and fine-tune to (smaller) downstream tasks.

For this, we remove the pre-trained prediction head and attach a zero-initialized D × K feedforward
layer, where K is the number of downstream classes

models trained extensively on large amounts of data to solve generic problems - typically selfsupervised (i.e. trained as generative AI to reproduce input data)

Foundational AI models

We use the ILSVRC-2012 ImageNet dataset with 1k classes
and 1.3M images, its superset ImageNet-21k with
21k classes and 14M images and JFT with 18k classes and
303M high-resolution images.

Typically, we pre-train ViT on large datasets, and fine-tune to (smaller) downstream tasks.

For this, we remove the pre-trained prediction head and attach a zero-initialized D × K feedforward
layer, where K is the number of downstream classes

3 class classification problem

models trained extensively on large amounts of data to solve generic problems - typically selfsupervised (i.e. trained as generative AI to reproduce input data)

Foundational AI models

regression problem

We use the ILSVRC-2012 ImageNet dataset with 1k classes
and 1.3M images, its superset ImageNet-21k with
21k classes and 14M images and JFT with 18k classes and
303M high-resolution images.

Typically, we pre-train ViT on large datasets, and fine-tune to (smaller) downstream tasks.

For this, we remove the pre-trained prediction head and attach a zero-initialized D × K feedforward
layer, where K is the number of downstream classes

Transfer Learning

can you use it for another task?

you have a model which was trained on some data

DOMAIN ADAPTATION: learning a model from a source data distribution and applying that model on a target data with a different distribution: the features are the same but have different distributions

e.g. Learn an energy model in one city (using building size, usage, occupancy) then apply it to a different city

?

Transfer Learning

does the model generalize to answer question on the new dataset with accuracy?

YES

NO

No need for additional learning: the model is transferable!

Fine Tune your model on the new data

you have a model which was trained on some data

Transfer Learning

What problems does it solve?

Small labelled dataset for supervised learning: use a model trained on a larger related dataset (and possibly fine tune with small amount of labels)

Limited computational resources because more are not available or to limit environmental impact of AI, as low level learning can be reused

 knowledge learned from a task is re-used in order to boost performance on a related task.

you have a model which was trained on some data

Transfer Learning

What problems does it solve?

Small labelled dataset for supervised learning: use a model trained on a larger related dataset (and possibly fine tune with small amount of labels)

Limited computational resources because more are not available or to limit environmental impact of AI, as low level learning can be reused

 knowledge learned from a task is re-used in order to boost performance on a related task.

you have a model which was trained on some data

Industry models like Chat-GPT or SAM are trained on huge amount of data we scientists could not afford to get!

Transfer Learning

What problems does it solve?

Small labelled dataset for supervised learning: use a model trained on a larger related dataset (and possibly fine tune with small amount of labels)

Limited computational resources because more are not available or to limit environmental impact of AI, as low level learning can be reused

 knowledge learned from a task is re-used in order to boost performance on a related task.

you have a model which was trained on some data

And large companies like Open-AI, Facebook, Google have unmatched computational resources

Fine-tuning by  retraining  everything for few epochs (few    10)

Start with the saved trained model:

weights and biases are set in the pre-trained model by training on Data 1

restart training from those weights and biases and adjust weights by running only a few epochs

prediction "head"

original data

fine tuning

Remember the "Deep Dream" demo and assignment

prediction "head"

early layers learn simple generalized features (like lines for CNN)

original data

fine tuning

Remember the "Deep Dream" demo and assignment

early layers learn simple generalized features (like lines for CNN)

prediction "head"

original data

fine tuning

late layers learn complex aggregate specialized features

fine tuning

Remember the "Deep Dream" demo and assignment

early layers learn simple generalized features (like lines for CNN)

prediction "head"

original data

late layers learn complex aggregate specialized features

Limited training - fine tuning

Remember the "Deep Dream" demo and assignment

Retrain (late layers and) head 

Replace input

prediction "head"

- Start with the weights as trained on the original dataset
- Train for a few epochs (sometimes as few as 10!)
The issue of vanishing gradient persists, but in this case it's helpful as it means we are mostly training the specialized layers at the end of the NN structure
- Makes large models accessible even if each training epoch is expensive by limiting the number of training epochs needed
- All rules of training need to be respected, including checking loss, adjusting learning rate, batch size (appropriately to the new dataset) etc

late layers learn complex aggregate specialized features

"Chop the head" fine tuning

Remember the "Deep Dream" demo and assignment

Replace input

early layers learn simple generalized features (like lines for CNN)

prediction "head"

late layers learn complex aggregate specialized features

"Chop the head" fine tuning

Remember the "Deep Dream" demo and assignment

"Freeze" early layers

Replace input

prediction "head"

late layers learn complex aggregate specialized features

"Chop the head" fine tuning

Remember the "Deep Dream" demo and assignment

"Freeze" early layers

Retrain (late layers and) head 

Replace input

prediction "head"

Can also modify the prediction head to change the scope of the NN (e.g. from classification to regression)

Fine-tuning how to:

https://www.tensorflow.org/guide/keras/transfer_learning

Fine-tuning how to:

parameter.requires_grad = False

(some models are really only available in pytorch ATM)

layer.trainable = False

Fine-tuning SAM example:

https://colab.research.google.com/drive/1takpjnf6R2bL9B4cyZZzFBkBEWTkfUGl?usp=sharing

loading a saved model

prints the number of parameters for every layer

Fine-tuning SAM example:

https://colab.research.google.com/drive/1takpjnf6R2bL9B4cyZZzFBkBEWTkfUGl?usp=sharing

checks if "gradient=true" i.e. if weights are trainable

Fine-tuning SAM example:

https://colab.research.google.com/drive/1takpjnf6R2bL9B4cyZZzFBkBEWTkfUGl?usp=sharing

checks if "gradient=true" i.e. if weights are trainable

sets gradient to false i.e. freezes the layer

Fine-tuning SAM example:

https://colab.research.google.com/drive/1takpjnf6R2bL9B4cyZZzFBkBEWTkfUGl?usp=sharing

Fine-tuning SAM example:

sets gradient to false i.e. freezes the layer

... only the "head" is left to be trainable