Giacomo Tarroni

Senior Lecturer in AI at City, University of London. Honorary Research Fellow at Imperial College London. Medical image analysis & machine learning.

[Giacomo Tarroni @g_tarroni]

We’ll be presenting FedCLAM in Daejeon, Korea:
Spotlight presentation — Wed 24 Sept, afternoon
Poster session — Thu 25 Sept, afternoon
Come by if you’re interested!

Full paper/code here: https://t.co/x40SZrNa4o

(3/3)

[Giacomo Tarroni @g_tarroni]

Our paper FedCLAM has been accepted at @MICCAI_Society 2025! 🎉
Another work led by Vasilis Siomos at @CityStGeorges focusing on FL in medical image segmentation.

Main novelties are 1) Client Adaptive Momentum and 2) Foreground Intensity Matching:

(1/3)

[Giacomo Tarroni @g_tarroni]

1) A momentum-based update rule that prioritises updates from clients learning effectively.
2) A loss function that aligns the intensity distributions of the predicted and ground truth foreground, preventing the model from learning scanner-specific artefacts.

(2/3)

[Giacomo Tarroni @g_tarroni]

ANFR consistently delivers top‐tier performance regardless of aggregation scheme, supports both global and personalised FL, and imposes only minimal overhead. And it trains like a charm!

Read more here: https://t.co/IJbA1vnyVj

(2/2)

[Giacomo Tarroni @g_tarroni]

Very happy to share that our recent work on Federated Learning (led by Vasilis Siomos at @CityStGeorges) has been published at @TmlrOrg!

It features ANFR, a simple architecture build specifically for FL, designed to tackle data heterogeneity directly at the client level. (1/2)

[Giacomo Tarroni @g_tarroni]

Dear @icmlconf, our recent submission received overwhelmingly positive reviews yet was rejected based on the AC’s factually incorrect evaluation, ignoring all reviews and rebuttal. We reached out via email to no avail.

What recourse do authors have in such situations? #ICML2025

[Giacomo Tarroni @g_tarroni]

[Please share!] We are offering 2x full PhD scholarships in AI/ML for cardiovascular research.
These are EPSRC-funded, 4-year scholarships reserved to candidates with Home/UK status. Reach out if interested! For more info:

https://t.co/sylylgJhzV

https://t.co/vuFqcMw3Qz

[Giacomo Tarroni @g_tarroni]

Paper & code for our unsupervised anomaly detection technique: https://t.co/CXxVmEjnWK

Paper & code for our FETS entry: coming up, stay tuned! (2/2)

[Giacomo Tarroni @g_tarroni]

Great times at #MICCAI2024 in Marrakech!

See Vasilis Siomos, PhD candidate at @Cit_AI1, doing a great job at presenting our work on Unsupervised Anomaly Detection via cold diffusion (accepted at the main conf) and our entry to the FETS challenge on Federated Learning (1/2)

[Giacomo Tarroni @g_tarroni]

Code repo linked in the pre-print above.

If interested, please feel free to reach out, either online or in person at MICCAI.

Thanks to all our co-authors: Vasilis Siomos, @MertBTwit, @ctzelepis, @BernhardKainz1

#AI #medicalimaging #unsupervisedlearning

[Giacomo Tarroni @g_tarroni]

Pushing unsupervised anomaly detection to greater heights!

Very happy to announce that our latest paper in the field, DISYRE v2, has been accepted for publication at #MICCAI2024.

Pre-print: https://t.co/CXxVmEjnWK

Work led by Sergio Naval Marimont @snavalm @Cit_AI1.

1/N:

[Giacomo Tarroni @g_tarroni]

DISYRE v2 outperforms all competitors in our experiments, and achieves +10% DICE over v1 on BraTS-T1 brain MR images!

See here some examples of test images, "healed" restorations and ensembled anomaly scores (grayscale) vs ground truth (red outline).

[Giacomo Tarroni @g_tarroni]

The upcoming merger between St George's and City, University of London will create a new Medical School and open up many interesting opportunities. In particular, the appointed lecturer will work in collaboration with the CitAI Research Centre @Cit_AI1 (and yours truly).

[Giacomo Tarroni @g_tarroni]

Sharing an opening for a Senior Lecturer in AI at the Cardiovascular and Genomics Research Institute at @StGeorgesU (merging with @CityUniLondon this August). If you have a background in AI/ML for medical/cardiac data, apply!

More info: https://t.co/uh1oljWL2a.

Please share!

[Giacomo Tarroni @g_tarroni]

Looking for a #PhD student to work on Weakly-Supervised Machine Learning for Medical Image Analysis at @CityUniLondon and @StGeorgesUni! Competitive bursary, fees fully covered for Home students (partial coverage for Overseas ones).

Please share!

https://t.co/tCfJkcM9zE

[Giacomo Tarroni @g_tarroni]

Please come and say hi if you'll be there. Congratulations to all co-authors!

Pre-prints:
ARIA: https://t.co/QiSuJsZcQf
DISYRE: https://t.co/4d7cYJbMeY

[Giacomo Tarroni @g_tarroni]

Very excited that both DISYRE (Diffusion-Inspired SYnthetic REstoration for Unsupervised Anomaly Detection) and ARIA (Interaction between Architectures, Aggregation methods and Initializations in federated visual classification) have been accepted for presentation at ISBI 2024!

[Giacomo Tarroni @g_tarroni]

Scholarship co-funded by @CityUniLondon and @CosmoIMD, global leaders in #AI for #medical #imaging with ground-breaking FDA-approved AI tools for medical video analysis. The candidate will join the @Cit_AI1 research centre under my supervision. Reach out if you have questions!

[Giacomo Tarroni @g_tarroni]

Hiring a full-time #PhD student with *full fee waiver* for Home/Overseas applicants to work on *Deep Learning for Medical Video Analysis and Understanding*. All info at the link below. Please share with your networks!

https://t.co/moPtHHrUHZ