Daniel Mbiye

Software Engineer - Interests: Machine Learning. Causal Inference. On-device ML. Autonomous Control. Robotics. Artificial Intelligence. Control Engineering.

[Daniel Mbiye @daniel_mbiye]

@fchollet Same!

[Andreas Orthey @andreas_orthey]

The last three years I gave a lecture series on "Motion Planning" together with Prof. Wolfgang Hönig (https://t.co/bvuK6TPTmx) at the @TUBerlin @LIS_TUBerlin.

We designed this course almost entirely from scratch to really capture the state of the art in this field, both from a research and from an industrial perspective.

I'm happy to share that you can now find this course for free on youtube. Please enjoy the content and find out how you can efficiently and optimally move any robot from A to B.

Website: https://t.co/P4ZsNlCSPf
Youtube Playlist: https://t.co/zu5pmIygoJ

Outline
Lecture 01: Organization, Introduction, Problem Formulation
Lecture 02: Transformations, Angular representations, Metrics, Efficient collision checking
Lecture 03: Graph-based Planning: Representations, A*, Admissible heuristics
Lecture 04: Advanced Search-Based Motion Planning
Lecture 05: Sampling-Based Geometric Motion Planning: PRMs
Lecture 06: Tree-based and Asymptotically-Optimal Planning
Lecture 07: Kinodynamic Planning: kinodynamic RRT, SST*, AO-x Geometric Planning: RRT-Connect, EST, PRM*
Lecture 08: Introduction to the Open Motion Planning Library (OMPL)
Lecture 09: Sampling-Based Motion Planning: More Theory and Planners (EST, RRT-Connect, PRM*, LazyPRM, FMT*); Intro to Optimization
Lecture 10: Optimization-Based Motion Planning
Lecture 11: Differential Flatness and SCP
Lecture 12: Optimization Wrap-Up and Method Comparison
Lecture 13: Multi-robot motion planning

[Kevin Patrick Murphy @sirbayes]

I am happy to announce that the first draft of my RL tutorial is now available.
https://t.co/SjMdabl0yW

[Andrej Karpathy @karpathy]

The (true) story of development and inspiration behind the "attention" operator, the one in "Attention is All you Need" that introduced the Transformer. From personal email correspondence with the author @DBahdanau ~2 years ago, published here and now (with permission) following some fake news about how it was developed that circulated here over the last few days.

Attention is a brilliant (data-dependent) weighted average operation. It is a form of global pooling, a reduction, communication. It is a way to aggregate relevant information from multiple nodes (tokens, image patches, or etc.). It is expressive, powerful, has plenty of parallelism, and is efficiently optimizable. Even the Multilayer Perceptron (MLP) can actually be almost re-written as Attention over data-indepedent weights (1st layer weights are the queries, 2nd layer weights are the values, the keys are just input, and softmax becomes elementwise, deleting the normalization). TLDR Attention is awesome and a *major* unlock in neural network architecture design.

It's always been a little surprising to me that the paper "Attention is All You Need" gets ~100X more err ... attention... than the paper that actually introduced Attention ~3 years earlier, by Dzmitry Bahdanau, Kyunghyun Cho, Yoshua Bengio: "Neural Machine Translation by Jointly Learning to Align and Translate". As the name suggests, the core contribution of the Attention is All You Need paper that introduced the Transformer neural net is deleting everything *except* Attention, and basically just stacking it in a ResNet with MLPs (which can also be seen as ~attention per the above). But I do think the Transformer paper stands on its own because it adds many additional amazing ideas bundled up all together at once - positional encodings, scaled attention, multi-headed attention, the isotropic simple design, etc. And the Transformer has imo stuck around basically in its 2017 form to this day ~7 years later, with relatively few and minor modifications, maybe with the exception better positional encoding schemes (RoPE and friends).

Anyway, pasting the full email below, which also hints at why this operation is called "attention" in the first place - it comes from attending to words of a source sentence while emitting the words of the translation in a sequential manner, and was introduced as a term late in the process by Yoshua Bengio in place of RNNSearch (thank god? :D). It's also interesting that the design was inspired by a human cognitive process/strategy, of attending back and forth over some data sequentially. Lastly the story is quite interesting from the perspective of nature of progress, with similar ideas and formulations "in the air", with a particular mentions to the work of Alex Graves (NMT) and Jason Weston (Memory Networks) around that time.

Thank you for the story @DBahdanau !

[Frank Dellaert @fdellaert]

We are in the process of editing a SLAM handbook, to be published by Cambridge University Press, with many stellar contributors. Part 1 is available as an online draft for public comments. Help us find bugs/problems!

[Glitchbyte @0xglitchbyte]

You can literally build whatever you want

[Caleb Watney @calebwatney]

This is the best paper written so far about the impact of AI on scientific discovery

[Daniel Mbiye @daniel_mbiye]

Nice weekend reading!

[Judea Pearl @yudapearl]

The Christian Roots of Bayesian Statistics.
https://t.co/mMCR4vhyN0
I'm Jewish! But that's not why I am only half Bayesian: https://t.co/gqeQb2lfLJ

[Daniel Mbiye @daniel_mbiye]

@realDonaldTrump This man forced the GOP congresspeople to repeat that lie about the stolen election or lose their seat. No respect for any future republican president knowing that he/she was spineless and chickened out because of the threat of losing their seat if they don’t toe the line.

[Daniel Mbiye @daniel_mbiye]

@realDonaldTrump Very sad to see what the GOP has become. For the first time in history, the world is witnessing a US presidential candidate who refused to accept the results of the previous presidential election and has cultivated this idea of a stolen election for the past four years.

[Rohan Paul @rohanpaul_ai]

Great work across 281-pages

"Alice's Adventures in a Differentiable Wonderland" ✨

Brilliant introduction to deep neural networks and differentiable programming.

Key areas covered ↓↓

• Automatic differentiation, stochastic optimization, and activation functions in depth and related core concepts.

• Convolutional networks, transformers, and graph neural networks

• Mathematical foundations including linear algebra, gradients, and probability theory

• Discusses implementation details in PyTorch and JAX

• Topic like Bayesian neural networks and neural scaling laws

This text offers a fresh perspective by presenting neural networks as assemblies of differentiable building blocks, departing from traditional biological comparisons

[Rohan Paul @rohanpaul_ai]

A really nice 628 page textbook for Mathematics from Carnegie Mellon University.

I like the Author's description

"The goal of this book is to help the reader make the transition from being a consumer of mathematics to a producer of it. This is what is meant by ‘pure’ mathematics. While a consumer of mathematics might learn the chain rule and use it to compute a derivative, a producer of mathematics might derive the chain rule from the rigorous definition of a derivative, and then prove more abstract versions of the chain rule in more general contexts (such as multivariate analysis).
Consumers of mathematics are expected to say how they used their tools to find their answers. Producers of mathematics, on the other hand, have to do much more: they must be able to keep track of definitions and hypotheses, piece together facts in new and interesting ways, and make their own definitions of mathematical concepts."

[quantmaxxer @quantmaxxer]

@Hamptonism Many such cases - Fematika (https://t.co/5jK47TY6qM) also has some absolute gems and then disappeared.

[Tianqi Chen @tqchenml]

The new semester is here, and one thing that excites me is working with fresh faces and having fun again on our deep learning systems course with @zicokolter, where we built a mini-PyTorch (needle) from scratch and use it to build deep learning models. https://t.co/EChWdUMVud

[Afshine Amidi @afshinea]

Announcing "Super Study Guide: Transformers & Large Language Models", a 250-page book with ~600 colored illustrations covering the concepts of the Stanford workshop that Shervine and I are teaching at this summer.

[Jeff Dean @JeffDean]

This book is really nicely illustrated! Nearly every page has multiple figures/diagrams that help explain the underlying concepts behind transformers & LLMs, including embeddings, attention, LoRA, distillation, quantization, ...

Nice work, @afshinea and @shervinea !

[Yann LeCun @ylecun]

Indeed, I do favor MPC over RL.
I've been making that point since at least 2016.
RL requires ridiculously large numbers of trials to learn any new task.
In contrast MPC is zero shot: If you have a good world model and a good task objective, MPC can solve new tasks without any task-specific learning.
That's the magic of planning.
It doesn't mean that RL is useless, but its use should be a last resort.

[Peter Stone @PeterStone_TX]

10 years after DQN, what are deep RL’s impacts on robotics? Which robotic problems have seen the most thrilling real-world successes thanks to DRL? Where do we still need to push the boundaries, and how?

Our latest survey explores these questions! Read on for more details. 👇

[Michael Pyrcz🌻 @GeostatsGuy]

Want to learn #MachineLearning? Do you appreciate super accessible educational content with all steps explained, excellent illustrations and well-documented, real examples?

I've shared my ML comprehensive, university course, complete with recorded lectures, codes, and workflows, to help you fast-track your #DataScience journey. Dive in and start learning today on #YouTube @ https://t.co/0nZzrXKgL6 ∀.

[Valeriy M., PhD, MBA, CQF @predict_addict]

“algorithms for decision making”, @MIT publishing book is freely available.