Ben Redhead

First year PhD student at the EPSRC CDT in Machine Learning Systems, Edinburgh University. Under the supervision of Prof. Amos Storkey and Prof. Victor Elvira.

[Ben Redhead @ben_redhead]

Happy to have started the Stars of Tomorrow internship at Microsoft Research Asia!

​Grateful to research large time-series models & novel generative architectures with my PI, Chang Xu. Big thanks to my CDT at @EdinburghUni for their support!

​#MSRA #ML #GenAI #TimeSeries

[Ben Redhead @ben_redhead]

Thrilled my Isambard AI grant proposal was accepted!

​Unlocking 10k GPU hrs on the UK's fastest supercomputer for Risk-aware Time-series Foundation Models.

​Huge thanks to my advisor and grant PI Amos Storkey & the MLSystems CDT at Edinburgh!

​#MachineLearning #IsambardAI

[Ben Redhead @ben_redhead]

@Jiankui_He Advancing science and medicine while potentially freeing babies from a life altering virus, nothing to apologize for.

[Ben Redhead @ben_redhead]

Fourier Analysis Networks: FAN

I would like to introduce a paper published utilizing an MLP architecture, switching a proportion of activation functions for sinusoidal functions. I have found this useful in Time Series Forecasting.

https://t.co/4alT5joJUa

#AI #MachineLearning

[Ben Redhead @ben_redhead]

@MIT_CSAIL I would like to work at @Microsoft, specifically Microsoft Research Asia to take part in their Time Series Forecasting research and learn more about developing robust learning algorithms in this area

[Ben Redhead @ben_redhead]

It is great to see the way the field is moving forward looking at generalisation and zero shot capabilities of foundation models for time series, I look forward to uploading the paper from myself and coauthors at PKU when it is ready!

[Ben Redhead @ben_redhead]

I had a great time at Microsoft Research Asia presenting my current research work in Non-Stationary Time Series Forecasting, it was great to interact with their research team and discuss current work in the field. I was also glad to be mentioned in their article in Chinese.

[Ben Redhead @ben_redhead]

Link to the article: https://t.co/kExCkGurEl

[Ben Redhead @ben_redhead]

@predict_addict @fjpjtwdm5r Please reference the independent review conducted by Microsoft Research Asia and HKUST which demonstrates competitive performance between the SOTA and foundation models. For convenience: https://t.co/8wFOtfIhS8

[Ben Redhead @ben_redhead]

@predict_addict @fjpjtwdm5r Just to add to this, I believe the best example is UniST which provides experimental results that improve over SOTA methods. https://t.co/8V71fGwPom by @GaoShanghua and @marinkazitnik

[Ben Redhead @ben_redhead]

@predict_addict @fjpjtwdm5r I provided the names of each model, the papers are available to review online. As specified these are specialised foundation models for time series not standard LLMs, I posted this in reply to you asking the other user what specialised meant in this context.

[Yann LeCun @ylecun]

Sometimes, the obvious must be studied so it can be asserted with full confidence:
- LLMs can not answer questions whose answers are not in their training set in some form,
- they can not solve problems they haven't been trained on,
- they can not acquire new skills our knowledge without lots of human help,
- they can not invent new things.
Now, LLMs are merely a subset of AI techniques.
Merely scaling up LLMs will *not* lead systems with these capabilities.
There is little doubt AI systems will have these capabilities in the future.
But until we have small prototypes of that, or at least some vague blueprint, bloviating about AI existential risk is like debating the sex of angels (or, as I've pointed out before, worrying about turbojet safety in 1920).

https://t.co/Qkyqo4WhO0

[Ben Redhead @ben_redhead]

@predict_addict @fjpjtwdm5r There are specialised foundation models for time series forecasting, such as TimesFM from Google Research, or if you want a model that takes a more similar approach to standard LLMs there is Chronos from AWS. Also UniST from Harvard and MIT outperforms SOTA forecasting models.

[nic lane @niclane]

At @flwrlabs our mission is to take #federatedlearning mainstream, and so we could not be more excited to be working w/ @DeepLearningAI to offer two new FL courses on core concepts, and applications to LLMs.

[Cambridge ML Systems Lab @CaMLSys]

We are delighted to announce the 8th PhD student to graduate from @CaMLSys: Yan Gao (@yangao381)! 🎆 Yan's PhD is one of the first to comprehensively study how SSL can federated (incl. video, speech, single-image). Yan was jointly supervised by @niclane7 and @parcolletT. 👏👏

[nic lane @niclane]

What are we going to do now that trends in model size (LLM and FMs) are breaking our ability to put them on-device? One answer is in the keynote from @SIVenieris @MobiUK2024 pitching the Consumer Edge 2.0 vision. Details in: https://t.co/jBDjGc5nBW to appear in @ieeepervasive

[nic lane @niclane]

"Say *no* to centralized training using *your* data!" says Javier (@jafermarq) of @flwrlabs today @MobiUK2024

[Ben Redhead @ben_redhead]

@jbhuang0604 This post is so insightful to my current situation. Thank you!

[Ben Redhead @ben_redhead]

@CapitalistPanda @SCMPNews Who gives you the right to say which declarations of diplomacy between countries are to be followed?
Pelosi was extremely selfish with this move and disregarded international diplomatic conventions