Ask HN: AI/ML papers to catch up with current state of AI? I used to be into ML back in the R-CNN, GAN, ResNet era and would read papers/blogs. Seems like ML is taking off recently and I want to get back into it! So far on my list I have attention is all you need, qlora, llama’s and q learning. Suggestions? |
just join communities on discord or locallama on reddit
Paper reference / main takeaways / link
instructGPT / main concepts of instruction tuning / https://proceedings.neurips.cc/paper_files/paper/2022/hash/b...
self-instruct / bootstrap off models own generations / https://arxiv.org/pdf/2212.10560.pdf
Alpaca / how alpaca was trained / https://crfm.stanford.edu/2023/03/13/alpaca.html
Llama 2 / probably the best chat model we can train on, focus on training method. / https://arxiv.org/abs/2307.09288
LongAlpaca / One of many ways to extend context, and a useful dataset / https://arxiv.org/abs/2309.12307
PPO / important training method / idk just watch a youtube video
Obviously these are specific to my work and are out of date by ~3-4 months but I think they do capture the spirit of "how do we train LLMs on a single GPU and no annotation team" and are frequently referenced simply by what I put in the "paper reference" column.
It's extremely zeitgeisty atm
Foundational model training got so expensive that unless you can get hired by "owns nuclear power plant of GPUs" you are not going to get any "research" done. And as the area got white-hot those companies have more available talent than hardware nowadays. So just getting into the practitioner area is the best way to get productive with those models. And you improve as a practitioner by practicing, not by reading papers.
If you're at the computer, your time is best spent writing code and interacting with those models in my opinion. If you cannot (e.g. commute) I listen to some stuff (e.g. https://www.youtube.com/watch?v=zjkBMFhNj_g - Anything from Karpathy on youtube, or https://www.youtube.com/@YannicKilcher channel).
This tool can help you find what's new & relevant to read. It's updated every day (based on ArXiv).
You can filter by category (Computer Vision, Machine Learning, NLP, etc), by release date, but most importantly, you can rank by PageRank (proxy of influence/readership), PageRank growth (to see the fastest growing papers in terms of influence), total # of citations, etc...
This is a great little book to take you from “vaguely understand neural networks” to the modern broad state of practice. I saw very little to quibble with. https://fleuret.org/francois/lbdl.html
It's a bit analogous to the situation with microprocessors. There is a ton of deep technical knowledge about how chips work, but most of this knowledge isn't critical for mainstream programming.
Read that first, then to keep up to date you can follow up with any papers that seem interesting to you. A good way to be aware of the interesting papers that come out is to follow @_akhaliq on X: https://twitter.com/_akhaliq
https://fleuret.org/francois/lbdl.html
I like that it’s formatted for the phone.
[1] https://simonwillison.net/2023/Aug/3/weird-world-of-llms/ [2] https://youtu.be/zjkBMFhNj_g?si=M6pRX66NrRyPM8x-
EDIT: Maybe I misunderstood as you asked about papers, not general intros. I don´t think that reading papers is the best way to "catch up" as the pace is rapid and knowledge very decentralized. I can confirm what Andrej recently wrote on X [3]:
"Unknown to many people, a growing amount of alpha is now outside of Arxiv, sources include but are not limited to:
- HN
- that niche Discord server
- anime profile picture anons on X
- reddit"
This is a good explanation of the Transformer details -> https://www.youtube.com/watch?v=bCz4OMemCcA&ab_channel=UmarJ...
This is old but covers a lot of background that you needs to know to understand very well the rest. What I like of this book is that it often explains in a very intuitive way the motivations behind certain choices. -> https://www.amazon.it/Natural-Language-Processing-Pytorch-Ap...
These papers outline the approach of reinforcement learning from human feedback which is being used to train lots of these LLMs such as ChatGPT.
Notwithstanding the above, I'd agree with others here who suggest learning by doing/implementing, not reading papers.
https://blog.oxen.ai/reading-list-for-andrej-karpathys-intro...
Above all, be wary of programmatic lists that claim to track the most important recent papers. There's a ridiculous amount of hype/propaganda and citation hacking surrounding new AI research, making it hard to discern what will truly stand the test of time. Tomas Mikolov just posted about this.[a]
---
http://bactra.org/notebooks/nn-attention-and-transformers.ht...
Definitely read through to the last section.
Here's also nice tour de building blocks, which could also double as transformers/tensorflow API reference documentation: https://www.youtube.com/watch?v=eMXuk97NeSI&t=207s
The #1 visualization of architecture and size progression: https://bbycroft.net/llm
From the past examples you give it sounds like you were into computer vision. There’s been a ton of developments since then, and I think you’d really enjoy the applications of some of those classic convolutional and variational encoder techniques in combination with transformers. A state of the art multimodal non-autoregressive neural net model such as Google’s Muse is a nice paper to work up to, since it exposes a breadth of approaches.
[2304.15004] Are Emergent Abilities of Large Language Models a Mirage? - arXiv https://arxiv.org/abs/2304.15004
Can't plan
https://openreview.net/forum?id=X6dEqXIsEW
No compositionality https://openreview.net/forum?id=Fkckkr3ya8
Apart from that it's great
https://www.thinkevolveconsulting.com/large-language-models-...
LLM (foundational papers)
* Attention is all you need - transformers + self attention
* BERT - first masked LM using transformers + self attention
* GPT3 - big LLM decoder (Basis of gpt4 and most LLM)
* Instruct GPT or TKInstruct (instruction tuning enables improved zero shot learning)
* Chain of Thought (improve performance via prompting)
some other papers which are become trendy depending on your interest
* RLHF - RL using human feedback
* Lora - make models smaller
* MoE - kind of ensembling
* self instruct - self label data
* constitutional ai - self alignment
* tree of thought - like CoT but a tree
* FastAttention,Longformer - optimized attention mechanisms
* React - agents
1. Transformers 2. Diffusion
The benefit is that, focus on understanding them both reeaaalllyy well and you are at the forefront of research;)
Also, what is the reason you want to do this? If it is about building some kind of AI enabled app, you don't have to read anything. Get an API key and let's go the barrier has never been lower.
https://arxiv.org/abs/2301.10743
Another interesting research topic is the trusted generation of tasks for finetuning
https://arxiv.org/abs/2306.08568
And I suppose too running these at the edge is terribly interesting too, if you can find analyses of "quantization" this is a highly active research are, and results are pretty incredible since it cuts resources by huge factors and no one quite knows why.
This is one that's easy to dive into with consumer hardware, but don't know any great papers myself
Run locally: https://github.com/ggerganov/llama.cpp
Quantized models: https://huggingface.co/TheBloke
Explainability is under research, though I haven't seen any good solutions.
This nay arise from skeptics who are calling the things stochastic parrots, incapable of reason, without a world model, etc.
I'd argue that there are plenty of less sexy, non-unicorn uses for AI/ML - particularly in industrial applications. SVMs, DNNs, etc are still very relevant. As is GOFAI in some domains.
Related question: how can I learn how to read the mathematical notation used in AI/ML papers? Is there a definitive work that describes the basics? I am a post-grad Engineer, so I know the fundamentals, but I'm really struggling with a lot of the Arxiv papers. Any pointers hugely appreciated.
If you find a sample, it may include the index of symbols in the beginning which is pretty comprehensive and may satisfy your question on its own.
https://www.goodreads.com/book/show/15857489-machine-learnin...
[0]: https://medium.com/@eric.christopher.ness/get-an-explanation...
>> To catch up with the current state of Artificial Intelligence and Machine Learning, it's essential to look at the latest and most influential research papers. Here are some categories and specific papers you might consider:
1. *Foundational Models and Large Language Models*: - Papers on GPT (Generative Pre-trained Transformer) series, particularly the latest like GPT-4, which detail the advancements in language models. - Research on BERT (Bidirectional Encoder Representations from Transformers) and its variants, which are pivotal in understanding natural language processing.
2. *Computer Vision*: - Look into papers on Convolutional Neural Networks (CNNs) and their advancements. - Research on object detection, image classification, and generative models like Generative Adversarial Networks (GANs).
3. *Reinforcement Learning*: - Papers from DeepMind, like those on AlphaGo and AlphaZero, showcasing advances in reinforcement learning. - Research on advanced model-free algorithms like Proximal Policy Optimization (PPO).
4. *Ethics and Fairness in AI*: - Papers discussing the ethical implications and biases in AI, including work on fairness, accountability, and transparency in machine learning.
5. *Quantum Machine Learning*: - Research on the integration of quantum computing with machine learning, exploring how quantum algorithms can enhance ML models.
6. *Healthcare and Bioinformatics Applications*: - Papers on AI applications in healthcare, including drug discovery, medical imaging, and personalized medicine.
7. *Robotics and Autonomous Systems*: - Research on the intersection of AI and robotics, including autonomous vehicles and drone technology.
8. *AI in Climate Change*: - Papers discussing the use of AI in modeling, predicting, and combating climate change.
9. *Interpretable and Explainable AI*: - Research focusing on making AI models more interpretable and explainable to users.
10. *Emerging Areas*: - Papers on new and emerging areas in AI, such as AI in creative arts, AI for social good, and the integration of AI with other emerging technologies like the Internet of Things (IoT).
To find these papers, you can check academic journals like "Journal of Machine Learning Research," "Neural Information Processing Systems (NeurIPS)," and "International Conference on Machine Learning (ICML)," or platforms like arXiv, Google Scholar, and ResearchGate. Additionally, following key AI research labs like OpenAI, DeepMind, Facebook AI Research, and university research groups can provide insights into the latest developments.