Top Calendar Slides Readings


Tech L170

Class Day/Time

MON 4pm - 7pm


Bryan Pardo Office hours by appointment

Course Description

WARNING: This course is about reading papers. A lot of papers. If you don’t want to read and discuss papers, this is not the course for you.

DESCRIPTION: One of the most exciting areas of research in deep learning is that of generative models. This course is dedicated to understanding the inner workings of the technologies that underlie these advances. In this TRANSFORMERS edition of the course, students will primarily learn about how Transformers + key auxiliary technologies (e.g. Variational Autoencoders, Reinforcement Learning) let them create text documents, write code, create music, images, speech, and more. We’ll also look at legal and ethical implications of these models. This is an advanced course that presumes a good working understanding of traditional supervised neural network technology and techniques (e.g. convolutional networks, LSTMs, loss functions, regularization, gradient descent).

PREREQUISITES: Course admission is by permission of instructor. The best Northwestern course to prepare for this class is CS 449 Deep Learning, although being a doctoral student in CS will likely get you in, even if you haven’t taken CS 449.

Course Calendar

Back to top

Week Day and Date Topic Presenter Commentators
  Mon Jan 8 Autoregressive language models Pardo  
    Attention Pardo  
    Transformers: The Illustrated Transformer Pardo  
  Mon Jan 22 Embeddings: The Illustrated Word2Vec Pardo  
    Positional Encoding Pardo  
    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding Alec P. Shang W. & Walker D.
  Mon Jan 29 Allocating Ownership Rights in Computer Generated Works Kelly J. Jon H. & Lirika S.
    Discussion of ethics and ownership with respect to generative models    
  Mon Feb 5 The Curious Case of Neural Text Degeneration Shubham S. Jon H. & Simon B.
    A Watermark for Large Language Models Charlotte L. Annie C. & Kelly J.
    Quantifying Memorization Across Neural Language Models Walker D. Shubham S. & Nathan G.
  Mon Feb 12 Deep Reinforcement Learning: Pong from Pixels Jon H. Natalia D. & James A.
    Deep reinforcement learning from human preferences Pardo Atmn P. & Izzy H.
    Training language models to follow instructions with human feedback Simon B. Lirika S & Shubham S.
  Mon Feb 19 Variational Auto Encoders (VAEs) Pardo  
    VQ-VAE: Neural Discrete Representation Learning James A. Adam H. & Nathan G.
    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale Shang W. Charlotte L. & Alec P.
  Mon Feb 26 Zero-Shot Text-to-Image Generation Atmn P. Conor K. & James A. & Alec P.
    MaskGIT: Masked Generative Image Transformer Natalia D. & Lirika S. Izzy H. & Simon B.
    Vampnet: Music Generation via Masked Acoustic Token Modeling Annie C. Walker D. & Negar K.
  Mon March 4 MusicLM: Generating Music From Text Izzy H. Annie C. & Natalia D.
    LORA: Low-rank Adaptation of Large Language Models Nathan G. Adam H. & Conor K.
    Diffusion/Score models Pardo  
  Mon March 11 Learning Transferable Visual Models From Natural Language Supervision Adam H. Charlotte L. & Shang W.
    GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models Conor K. Kelly J. & Atmn P.
    Hierarchical Text-Conditional Image Generation with CLIP Latents - -

Course assignments

Reading: 50 points

You will submit 25 one-page reviews of readings from the course website. 15 of these must be papers (not lecture slides, actual papers) scheduled for presenation in the course calendar. 10 of these can be chosen from the full set of papers for the course.

Class Paper Presentation: 30 points

Once during the course of the term, you (and your partner) will be the lead in discussing the reading in class. This will mean you haven’t just read the paper, but you’ve read related work, really understand it and can give a 30-minute presentation of the paper (including slides) and then lead a discussion about it.

Being a Paper Commentator: 20 points

For 2 papers OTHER than your own presented paper, you’ll be expected to be 100% on top of the material and be the counterpoint to the presenter’s point. I’ll expect you to be engaged in the discussion and to display clear knowledge of the material.

Lecture Slides and Notebooks

  1. Language Models

  2. Attention Networks

  3. Embeddings

  4. Positional Encoding, Attention and Embeddings Notebook

  5. Transformers

  6. Deep Reinforcement Learning: Q Learning

  7. Deep Reinforcement Learning: Policy Gradients

  8. Deep Reinforcement Learning: Human Feedback

  9. Variational Inference

  10. Variational Autoencoders

  11. Generative Adversarial Networks

  12. Diffusion/Score Models

Back to top


Course Reading

Back to top


  1. Pixel Recurrent Networks: A highly infulential autoregressive model for image generation

  2. WaveNet: A Generative Model for Raw Audio: A highly infulential autoregressive model for audio generation


Architectural elements that lead up to Transformers

  1. Visualizing A Neural Machine Translation Model (Mechanics of Seq2seq Models With Attention) This is a good starting point blog on attention models, which is what Transformers are built on.

  2. Sequence to Sequence Learning with Neural Networks: This is the paper that the link above was trying to explain.

  3. Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation: This introduces encoder-decoder networks for translation. Attention models were first built on this framework.

  4. Neural Machine Translation by Jointly Learning to Align and Translate: This paper introduces additive attention to an encoder-decoder.

  5. Effective Approaches to Attention-based Neural Machine Translation: This paper introduces multiplicative attention, which is what Transformers use.

  6. Deep Residual Learning for Image Recognition: This introduces the idea of “residual layers”, which are layers that are skippable. This idea is used in Transformers.


  1. The Illustrated Word2Vec: Transformers for text take word embeddings as input. So what’s a word embedding? This is a walk through word embeddings, at a high level, with no math.

  2. Efficient Estimation of Word Representations in Vector Space: This is the Word2Vec paper.

  3. GloVe: Global Vectors for Word Representation: The paper that describes the Glove embedding, which is an improvement on Word2Vec, and has downloadable embeddings to try. There is math here.

  4. Using the Output Embedding to Improve Language Models: In transformers, they actually learn their embeddings at the same time as everything else and tie the input embedding to the output embedding. This paper explains why.

The Transformer Architecture

  1. The Illustrated Transformer: A good initial walkthrough that helps a lot with understanding transformers ** I’d start with this one to learn about transformers.**

  2. The Annotated Transformer: An annotated walk-through of the “Attention is All You Need” paper, complete with detailed python implementation of a transformer. ** If you actually want to understand transformer implementation, you should read this in depth…and play with the code.**

  3. Attention is All You Need: The paper that introduced transformers, which are a popular and more complicated kind of attention network.

About positional encoding

  1. Self-Attention with Relative Position Representations: The most frequently used alternative to absolute positional encoding

  2. Rotary Positional Encoding: claims to combine the benefits of both absolute and relative positional encoding

BERT and GPT, two foundational architectures

  1. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding: A widely-used language model based on Transformer encoder blocks.

  2. The Illustrated GPT-2: A good overview of GPT-2 and its relation to Transformer decoder blocks.

  3. GPT-3:Language Models are Few-Shot Learners: This explores the range of things that you can do with a GPT model.

Why sampling strategies matter

  1. The Curious Case of Neural Text Degeneration: When you sample from the output of a language model, it matters a LOT just how you sample. Read this to understand why.

  2. A Watermark for Large Language Models: Can you use a sampling strategy to watermark generated text?

Reinforcement learning makes ChatGPT possible

  1. Training language models to follow instructions with human feedback: This combines RL with a GPT model to make InstructGPT, the precursor to ChatGPT

  2. Learning Transferable Visual Models From Natural Language Supervision: This describes how the original DALL-E selects which of the many images it generates should be shown to the user.

Symbolic Music Making with Transformers

  1. Music Transformer: Applying Transformers to music composition.

  2. Anticipatory Music Transformer: A Controllable Infilling Model for Music

Making Images with Transformers

  1. Image GPT: Using a Transformer to make images. This isn’t DALL-E, even though it’s by OpenAI.

  2. Zero-Shot Text-to-Image Generation: This is the original version of DALL-E, which generates images conditioned on text captions. It is based on Transformer architecture.

  3. MaskGIT: Masked Generative Image Transformer This lets you generate in PARALLEL, not auto-regressively

Making Transformers Efficient

  1. Reformer: The Efficient Transformer: This uses locality sensitive hashing to make attention much more efficient, moving it from taking O(n^2) and making it O(nlogn).

    Adding Memory/Scratch Space to a Transformer

  2. Memory Transformer

  3. Vision Transformers Need Registers

  4. Think before you speak: Training Language Models With Pause Tokens

  5. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Language Modeling in the Audio Domain

  1. WAV2VEC: UNSUPERVISED PRE-TRAINING FOR SPEECH RECOGNITION: This describes a way to build a dictionary of audio tokens that is used in MusicLM

  2. Wav2vec 2.0: Learning the structure of speech from raw audio: The 2nd iteration of wav2vec

  3. SoundStream: An End-to-End Neural Audio Codec: Perhaps the top (as of 2023) audio codec. It is used in multiple audio langague models to tokenize the audio for the language model.

  4. W2v-BERT: Combining Contrastive Learning and Masked Language Modeling for Self-Supervised Speech Pre-Training: Masked inference language model method to learn audio tokens. This is what is actually used in MusicLM

  5. AudioLM: A Language Modeling Approach to Audio Generation: A language model for generating speech continuation

  6. MusicLM: Generating Music From Text: A model generating music audio from text descriptions such as “a calming violin melody backed by a distorted guitar riff”.

  7. Vampnet: Music Generation via Masked Acoustic Token Modeling: Generating non-autoregressively, similar to MaskGit

Fine tuning of langauge models

  1. LORA: Low-rank Adaptation of Large Language Models: Microsoft’s approach to fast, efficient retraining for downstream tasks.

Reinforcement learning for Language Models

  1. Reinforcement Learning: An Introduction: This is an entire book, but it is the one I learned RL from.

  2. Policy Gradient Methods: Tutorial and New Frontiers: This is a video lecture that explains reinforcement learning policy grading methods. This is underlying tech used for training ChatGPT. Yes, this video is worth a “reading” credit. Yes, I started it 37 minutes into the lecture on purpose. You don’t have to watch the first half of the lecture.

  3. Andrew K.s blog on Deep Reinforcement Learning: When combined with the video tutorial above, you’ll more-or-less understand policy gradient methods for deep reinforcement learning

  4. Proximal Policy Optimization Algorithms: The paper that (mostly) explains the RL approach used in InstructGPT (the precursor to ChatGPT)

  5. This blog on RL from human feedback: read the paper linked at the start of the blog. It teaches how to learn a reward function from human feedback, so you can do RL.

  6. This blog on Aligning language models to follow instructions together explain how ChatGPT is fine-tuned to do prompt answering by combining proximal policy optimization and RL from human feedback (the two previous papers on this list).

  7. ChatGPT: Already perhaps the most famous chatbot and most famous language model and it has been out about 2 weeks as of this writing.

Multimodal Langauge Modeling

  1. GPT-4 with vision (GPT-4V): enables users to instruct GPT-4 to analyze image inputs provided by the user, and is the latest capability we are making broadly available

  2. LLaVA: Large Language and Vision Assistant:a large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding

  3. LLark: A Multimodal Foundation Model for Music: Mix Jukebox and Llama 2 and you get this.


Creating adversarial examples

  1. Explaining and Harnessing Adversarial Examples : This paper got the ball rolling by pointing out how to make images that look good but are consistently misclassified by trained deepnets.

Creating GANs

  1. Generative Adversarial Nets: The paper that introduced GANs

  2. 2016 Tutorial on Generative Adversarial Networks by one of the creators of the GAN. This one’s long, but good.

  3. DCGAN: Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks: This is an end-to-end model. Many papers build on this.

Advanced GANS

  1. PROGRESSIVE GROWING OF GANS FOR IMPROVED QUALITY, STABILITY, AND VARIATION This is used in StyleGAN and was state-of-the-art in 2018.

  2. StyleGAN: A Style-Based Generator Architecture for Generative Adversarial Networks: As of 2019, this was the current state-of-the-art for GAN-based image generation.

  3. StyleGAN2-ADA: Training Generative Adversarial Networks with Limited Data: : As of 2020, this was the current state-of-the-art for GAN-based image generation.

  4. Cross-Modal Contrastive Learning for Text-to-Image Generation As of 2021, this was the best GAN for text-conditioned image generation. Note it’s use of contrastive loss. You’ll see that again in CLIP.

  5. Adversarial Audio Synthesis: introduces WaveGAN, a paper about applying GANs to unsupervised synthesis of raw-waveform audio.

  6. MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis: Doing speech synthesis with GANs.

  7. HiFi-GAN: Generative Adversarial Networks for Efficient and High Fidelity Speech Synthesis: Even better speech synthesis with GANs.

Variational Auto Encoders (VAEs)

Background needed for Variational Autoencoders

  1. The Deep Learning Book’s Chapters on Probability and Linear Algebra. Read these before the Easy Intro to KL Divergence

  2. An Easy Introduction to Kullback-Leibler (KL) Divergence . Read this before reading about ELBO

  3. Jenson’s Inequality (an example with code). Read this before reading about ELBO

  4. Evidence Lower Bound, Clearly Explained: A video walking through Evidence Lower Bound.

  5. A walkthrough of Evidence Lower Bound (ELBO): Lecture notes from David Blei, one of the inventors of ELBO. ELBO is what you optimize when you do variational inference in a VAE.

  6. Categorical Reparameterization with Gumbel-Softmax: This is a way of allowing categorical latent variables in your model so you can run a differentiable gradient descent algorithm through them. This is used in Vector-Quantized VAEs.

  7. Probabilistic Graphical Models: Lecture notes from the class taught at Stanford.


  1. A starter blog on AutoEncoders and VAEs: Probably a good place to start.

  2. The Deep Learning Book’s Chapter on Autoencoders

  3. From neural PCA to deep unsupervised learning : This paper introduces Ladder networks, which will come back when we get to VAEs

BASIC Variational Auto Encoders (VAEs)

  1. Tutorial on Variational Autoencoders: This is a walk-through of the math of VAEs. I think you should maybe start with this one.

  2. Variational Inference, a Review for Statisticians: This explains the math behind variational inference. One of the authors is an inventor of variational inference.

  3. An introduction to variational autoencoders: This is by the inventors of the VAE.

  4. Tutorial: Deriving the Standard Variational Autoencoder (VAE) Loss Function: This is the only paper I’ve found that walks you through all the details to derive the actual loss function.


  1. VQ-VAE: Neural Discrete Representation Learning

  2. Conditional VAE: Learning Structured Output Representation using Deep Conditional Generative Models: Making a controllable VAE through conditioning

  3. Beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework : This is about making disentangled representations: making the VAEs latent variables meaningful to us.

  4. Isolating Sources of Disentanglement in VAEs: More on disentangled representations in VAEs

  5. Ladder VAEs: Hierarchical VAEs

  6. Adversarial Auto-Encoders: You can guess what this is.

  7. A Wizard’s Guide to Adversarial Autoencoders: This is a multi-part tutorial that will be helpfup for understanding AAEs.

  8. From Autoencoder to Beta-VAE: Lilian Weng’s overview of most kinds of autoencoders

VAE Applications

  1. MUSIC VAE: Learning Latent Representations of Music to Generate Interactive Musical Palettes: Making controllable music composition with VAEs

  2. Jukebox: A Neural Net that Generates Music….with a combination of autoencoders and GANs

  3. Accelerated antimicrobial discovery via deep generative models and molecular dynamics simulations: Using a WAE to generate new drugs

Diffusion and Score Models

  1. Deep unsupervised learning using nonequilibrium thermodynamics: The 2015 paper where diffusion models were introduced.

  2. Denoising Diffusion Probabilistic Models: This was a break-out paper from 2020 that got people excited about diffusion models.

  3. Generative Modeling by Estimating Gradients of the Data Distribution: This is a blog that explains how score-based models are also basically diffusion models.

  4. What are Diffusion Models?

  5. An Introduction to Diffusion Models: A nice tutorial blog that has Pytorch code.

Advanced Difussion and Score Models

  1. Guidance: a cheat code for diffusion models: if you want to understand DALL-E-2 and Imagen, you need to understand this.

  2. DiffWave: A Versatile Diffusion Model for Audio Synthesis: A neural vocoder done with diffusion from 2021

  3. Universal Speech Enhancement With Score-based Diffusion

  4. Cold Diffusion: Inverting Arbitrary Image Transforms Without Noise: Do we need to add noise at each step or would any transform do?

  5. Consistency Trajectory Models. For single-step diffusion model sampling, our new model, Consistency Trajectory Model, achieves SOTA on CIFAR-10 (FID 1.73) and ImageNet 64x64 (FID 1.92). CTM offers diverse sampling options and balances computational budget with sample fidelity effectively.

  1. High Fidelity Image Generation Using Diffusion Models: A Google Blog that gives the chain of development that led to Imagen.

  2. Google’s Imagen: This is the Pepsi to DALL-E-2’s Coke.

  1. Diffusion Models Beat GANs on Image Synthesis: This paper describes many technical details used in the GLIDE paper…and therefore in DALL-E-2

  2. GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models: This is paper that lays the groundwork for DALL-E-2.

  3. Learning Transferable Visual Models From Natural Language Supervision: The CLIP representation. This is used in DALL-E-2.

  4. Hierarchical Text-Conditional Image Generation with CLIP Latents: The DALL-E-2 paper.

Ethics and Societal Effects of Generative Modeling

  1. Pam Samuelson’s AI Meets Copyright. This is a video lecture on generative AI and copyright law from one of top copyright scholars in the USA.

  2. Allocating Ownership Rights in Computer Generated Works

  3. Sociotechnical Safety Evaluation of Generative AI Systems: A big overview paper on how to ensure your generative AI is going to minimize possible harms.

  4. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?: This is the paper that Timnit Gebru and Margaret Mitchell got fired from Google’s Ethical AI team for publishing.

  5. Alignment of Language Agents: This is Deep Mind’s critique of their own approach.

  6. Open AI’s analysis of GPT-4 potential harms: Worth a serious read

  7. The Ethical Implications of Generative Audio Models: A Systematic Literature Review: By Northwestern’s own Julia Barnett!

How bad is the memorization problem in generative models?

  1. Quantifying Memorization Across Neural Language Models: Systematic experiments on how model size, prompt length, and frequency of an example in the training set impact our ability to extract memorized content.

  2. Extracting Training Data from Large Language Models: Did GPT-2 memorize a Harry Potter book? Read this and find out.

  3. Extracting Training Data from Diffusion Models: Exactly what it sounds like.

  4. Diffusion Art or Digital Forgery? Investigating Data Replication in Diffusion Models: Just what it sounds like.


Normalizing Flows

  1. Variational Inference with Normalizing Flows: A differentiable method to take a simple distribution and make it arbitrarily complex. Useful for modeling distributions in deep nets. Can be added to VAEs. These are being replaced by diffusion models.

FILM Layers

  1. FiLM: Visual Reasoning with a General Conditioning Layer: Affine transformation of input layers that proves helpful in many contextx. Here’s the TL;DR version. I’d start with the TL;DR.

Structured State Space Models

  1. Efficiently Modeling Long Sequences with Structured State Spaces

  2. The Annotated S4: This is a guided walk through (with code) of a structured state space model.

  3. It’s Raw! Audio Generation with State-Space Models

  4. Mamba: Linear-Time Sequence Modeling with Selective State Spaces: Adds input-dependent “attention” (selectivity) to SSMs and claims to beat Transformers at their own game.


  1. Competition-level code generation with AlphaCode: This beats 1/2 of all human entrants into a coding competition.

  2. Toolformer: Language Models Can Teach Themselves to Use Tools: Meta researchers claim a transformer can learn to use APIs.