PhD Student @ University of Edinburgh
Joined March 2017
- Tweets 159
- Following 1,266
- Followers 526
- Likes 4,491
29 Photos and videos
Pinned Tweet
7 Oct 2024
We learn more expressive mixture models that can subtract probability density by squaring them
๐จWe show squaring can reduce expressiveness
To tackle this we build sum of squares circuits๐
๐We explain why complex parameters help, and show an expressiveness hierarchy around๐
3
32
163
21,459
Lorenzo Loconte retweeted
Mar 29
AI people going on about Johnson-Lindenstrauss lemma like it was discovered yesterday.
itโs just another example of how most folks donโt read or know anything more than 5 years old
9
11
308
25,838
Lorenzo Loconte retweeted
Happy to share a major milestone: after years of development, we are officially launching Version 1.0 of the GeometricKernels library!
To top it off, our accompanying paper has just been published in JMLR (MLOSS)! ๐
github.com/geometric-kernelsโฆ
7
108
739
34,199
Lorenzo Loconte retweeted
Feb 12
I am a bit late to the party, but I am happy to share that our latest work was accepted to #ICLR2026 ๐ฅณ๐ฅณ
๐ How to Square Tensor Networks and Circuits Without Squaring Them
arxiv.org/abs/2512.17090
1
1
10
335
Lorenzo Loconte retweeted
28 Dec 2025
Good news everyone! This year we will be organizing a workshop on Unifying Concept Representation Learning at ICLR'26!
The workshop is about unifying ideas and techniques from #NeSy AI, #XAI and #Causal representation learning. Have a look at the CfP!
28 Dec 2025
๐ข Announcing the Workshop on **Unifying Concept Representation Learning** at ICLRโ26 ( @iclr_conf ).
When? 26 or 27 April 2026
Where? Rio de Janeiro, Brazil
Call for papers, schedule, invited speakers & more: ucrl-iclr26.github.io
Looking forward to your submissions!
2
18
25
3,370
Lorenzo Loconte retweeted
19 Sep 2025
Congratulations to everyone who got their @NeurIPSConf papers accepted ๐๐๐
At #EurIPS we are looking forward to welcoming presentations of all accepted NeurIPS papers, including a new โSalon des Refusรฉsโ track for papers which were rejected due to space constraints!
4
25
238
51,339
We're glad to announce the NeSy 2025 Test of Time award for "Probabilistic Inference Modulo Theories"!
๐Rodrigo de Salvo Braz was here to accept the award.
This is groundwork for recent NeSy approaches like DeepSeaProbLog and the probabilistic algebraic layer.
1
8
13
890
Lorenzo Loconte retweeted
1 Sep 2025
Insightful! We tackled the same problem in Knowledge Graph Completion. Dot-product scoring on low-dim embeddings severely limits what a model can predict. We call this a โrank bottleneckโ to align with the existing LM literature. Our paper for context: arxiv.org/abs/2506.22271
1
2
4
389
Lorenzo Loconte retweeted
16 Jul 2025
EurIPS is coming! ๐ฃ Mark your calendar for Dec. 2-7, 2025 in Copenhagen ๐
EurIPS is a community-organized conference where you can present accepted NeurIPS 2025 papers, endorsed by @NeurIPSConf and #NordicAIR and is co-developed by @ELLISforEurope
eurips.cc
1
27
85
13,401
Lorenzo Loconte retweeted
10 Jul 2025
Spotlight poster coming soon at #ICML2025 @icmlconf!
๐East Exhibition Hall A-B E-1806
๐๏ธWed 16 Jul 4:30 p.m. PDT โ 7 p.m. PDT
๐arxiv.org/pdf/2410.12537
Letโs chat! Iโm always up for conversations about knowledge graphs, reasoning, neuro-symbolic AI, and benchmarking.
17 Oct 2024
๐จIs complex query answering really complex?๐จ
unfortunately not!
the current benchmarks boil down to link prediction 98% of the time...
how to fix this???
๐๐๐
๐arxiv.org/abs/2410.12537
with @c_gregucci @BoXiongs @loreloc_ @PMinervini @ststaab
5
20
4,101
Lorenzo Loconte retweeted
13 Jun 2025
๐งตWhy are linear properties so ubiquitous in LLM representations?
We explore this question through the lens of ๐ถ๐ฑ๐ฒ๐ป๐๐ถ๐ณ๐ถ๐ฎ๐ฏ๐ถ๐น๐ถ๐๐:
โAll or None: Identifiable Linear Properties of Next-token Predictors in Language Modelingโ
Published at #AISTATS2025๐ด
1/9
8
55
314
29,365
Lorenzo Loconte retweeted
21 May 2025
We propose Neurosymbolic Diffusion Models! We find diffusion is especially compelling for neurosymbolic approaches, combining powerful multimodal understanding with symbolic reasoning ๐
Read more ๐
32
105
589
54,332
Lorenzo Loconte retweeted
14 May 2025
Just under 10 days left to submit your latest endeavours in โก#tractableโก probabilistic modelsโ
Join us at TPM @auai.org #UAI2025 and show how to build #neurosymbolic / #probabilistic AI that is both fast and trustworthy!
16 Apr 2025
the #TPM โกTractable Probabilistic Modeling โกWorkshop is back at @UncertaintyInAI #UAI2025!
Submit your works on:
- fast and #reliable inference
- #circuits and #tensor #networks
- normalizing #flows
- scaling #NeSy #AI
๐ deadline: 23/05/25
๐ tractable-probabilistic-modeโฆ
4
6
638
Lorenzo Loconte retweeted
23 Apr 2025
In LoCo-LMs, we propose a neuro-symbolic loss function to fine-tune a LM to acquire logically consistent knowledge from a domain graph, i.e. wrt. to a set of logical consistency rules.
@looselycorrect @tetraduzione
arxiv.org/abs/2409.13724
1
4
12
2,723
Lorenzo Loconte retweeted
23 Apr 2025
We developed a library to make logical reasoning embarrassingly parallel on the GPU.
For those at ICLR ๐ธ๐ฌ: you can get the juicy details tomorrow (poster #414 at 15:00). Hope to see you there!
1
7
15
662
Lorenzo Loconte retweeted
23 Apr 2025
๐จNew at #ICLR: we introduce the first ever ๐ฅ๐๐ฒ๐๐ซย that makesย ๐๐ง๐ฒย neural networkย ๐๐จ๐ฆ๐ฉ๐ฅ๐ข๐๐ง๐ญ ๐๐ฒ ๐๐๐ฌ๐ข๐ ๐ง with constraints expressed asย ๐๐ข๐ฌ๐ฃ๐ฎ๐ง๐๐ญ๐ข๐จ๐ง๐ฌ ๐จ๐ ๐ฅ๐ข๐ง๐๐๐ซ ๐ข๐ง๐๐ช๐ฎ๐๐ฅ๐ข๐ญ๐ข๐๐ฌโeven if they defineย ๐ง๐จ๐ง-๐๐จ๐ง๐ฏ๐๐ฑ ๐ฌ๐ฉ๐๐๐๐ฌ!
1
7
35
3,263
Lorenzo Loconte retweeted
14 Mar 2025
hi, we found problematic benchmarks in complex query answering as well! arxiv.org/abs/2410.12537 @tetraduzione @Mniepert @c_gregucci @chrsmrrs @michael_galkin
4
14
1,073
Lorenzo Loconte retweeted
4 Mar 2025
Our paper "Low-rank finetuning for LLMs is inherently unfair" won a ๐๐๐ฌ๐ญ ๐ฉ๐๐ฉ๐๐ซ ๐๐ฐ๐๐ซ๐ at the @RealAAAI colorai workshop! #AAAI2025
Congratulations to amazing co-authors @nandofioretto @WatIsDas @CuongTr95450563 and M. Romanelli ๐ฅณ๐ฅณ๐ฅณ
31 May 2024
Fine-tuning your LLM with LoRA for critical areas like โ๏ธ criminal justice, ๐ฅ healthcare, or ๐ผ hiring?
โ ๏ธ Think again! โ ๏ธ
๐จ We found that LoRA can amplify #AI #harms:
โ๏ธFalse sense of #safety #alignment
๐ซIncreased #unfairness and #bias, hitting minority groups the hardest
4
3
39
4,044
4 Mar 2025
Live from the CoLoRAI workshop at AAAI
april-tools.github.io/coloraโฆ
@nadavcohen is now giving his talk on "What Makes Data Suitable for Deep Learning?"
Tools from quantum physics are shown to be useful in building more expressive deep learning models by changing the data distribution
1
1
6
520
4 Mar 2025
The last speaker of the workshop is Alexandros Georgiou, who is giving an introduction to polynomial networks and equivariant tensor network architecture, as well as how to implement them.
1
1
133
4 Mar 2025
You can find the speakers bios and the abstracts of presentations here: april-tools.github.io/coloraโฆ
Check them out!
1
105