@lorebasile

Postdoc @AreaSciencePark

 Trieste, Italy
Joined June 2020
  • Tweets19
  • Following332
  • Followers56
4Media
Just landed in San Diego to present "Head Pursuit: Probing Attention Specialization in Multimodal Transformers", our spotlight paper @NeurIPSConf! Don't miss poster #1013 on Wednesday, Dec 3 at 11AM. [1/6]
1
5
1
17
1,255
Lorenzo Basile retweeted
Berker Demirel@berkerdemirel97Oct 13
In JEPA, are we preventing collapse in the representation we actually use downstream? TLDR: not necessarily. A high-rank projected representation does not guarantee a non-collapsed backbone. λ-JEPA is designed to target this gap. arxiv.org/abs/2609.35288
7
64
6
465
38,552
Irene is presenting our paper LOCUS today at the Actionable Interpretability workshop at #COLM2026. We explore a more localized form of activation steering, selecting specific heads and directions based on the model’s unembedding. Don’t miss her poster at 10.50 if you're there!
Irene Tallini@karana_karanaOct 8
🎯 LOCUS: Head Selection and Subspace Projection for Targeted Activation Steering How do we steer model behavior without degrading performance? LOCUS uses token subspaces tied to the target property to choose which heads to steer, and which subspace within each head.
13
193
Check out DEVIL, now out in @NatureComms! DEVIL is a Bayesian, GPU‑powered framework for single‑cell differential expression that lets you analyze cohorts with millions of cells from multiple patients without losing statistical rigor. 📄doi.org/10.1038/s41467-026-7…
1
2
1
14
5,197
Happy to announce that one of my PhD papers is out! We build and analyse a synthetic WGS cohort of >200 samples with complete evolutionary ground truth: caravagnalab.github.io/SCOUT… Joint effort with @GiorgiaGandolf1 and thanks to all co-authors and supervisors! Thread below!
Giulio Caravagna@gcaravagnaJun 15
🧬 New preprint: SCOUT (Synthetic Cohort Of Universal Tumours) A benchmark for cancer WGS where the entire tumour evolutionary history is known. 📊 7 tumour models 🧪 24 biopsies 💾 223 FASTQs 🗄️ >27 TB of data! #CancerGenomics #tumourevolution caravagnalab.github.io/SCOUT…
2
7
607
Lorenzo Basile retweeted
Valentino Maiorca@ValeMaiorcaMay 21
🎨Activation steering can reliably push a text-to-image generator toward a visual concept, but at a cost: each concept needs its own estimation. ⚡HyperTransport (HT) predicts the intervention directly, matching per-concept SOTA at 3–4 orders of magnitude less cost. [1/6]
3
18
54
4,836
Steering generative models with LinEAS is: – parameter-efficient – fine-grained – data-efficient – robust across models and modalities Come check it out at #NeurIPS2025! 🗓️ Dec 5 • 4:30–7:30 PM PST 📍 Exhibit Hall C,D,E #3604
Pau Rodríguez@prlz7721 Oct 2025
🚀 Excited to share LinEAS, our new activation steering method accepted at NeurIPS 2025! It approximates optimal transport maps e2e to precisely guide 🧭 activations achieving finer control 🎚️ with ✨ less than 32 ✨ prompts! 💻github.com/apple/ml-lineas 📄arxiv.org/abs/2503.10679
6
12
1,646
Come find us at NeurIPS Poster Session 6 #1114 4PM - 7PM to hear how multimodal models develope localized image-text communication in their internal representations. #NeurIPS2025 @alexserra1998 @francescortu @lorebasile @DiegoDoimo
Alessandro Pietro Serra@aleserra19983 Dec 2025
Excited to present "The Narrow Gate: Localized Image-Text Communication in Native Multimodal Models" at #NeurIPS2025 and #EurIPS2025! 📍NeurIPS: Poster Session 6 #1114 (Fri Dec 5 4pm-7pm) 📍EurIPS: Poster Session D3 #88 (Wed Dec 3 10:30am-12:30pm) 🧵👇
2
7
521
🌈 Property enhancement/inhibition We can control image captioning with head-level interventions. Rescaling heads specialized on colors or sentiments we can promote or remove related words, without finetuning. [5/6]
1
3
148
Amazing collaboration with @ValeMaiorca, @DiegoDoimo, @FrancescoLocat8 and @albecazzaniga. 📘 Paper: arxiv.org/abs/2510.21518 💻 Code: github.com/lorenzobasile/Hea… @AreaSciencePark @SapienzaRoma @ISTAustria [6/6]
3
99
🤖Seminar Series Next week, Giorgos Nikolaou and Tommaso Mencattini (@EPFL) will present "Language Models are Injective and Hence Invertible". @GiorgosNik02 @tommaso_mncttn 📅Nov 27, 11 CET (Online) 👉Register to attend: bit.ly/4ieKRYq
6
1
11
1,213
Excited to share that 2/2 papers from our Lab (LADE) were accepted to #NeurIPS2025 (one spotlight 🎉) Great work from all the students and collaborators involved! @AreaSciencePark @aleserra1998 @lorebasile @francescortu @lucrevaleriani @DiegoDoimo @ansuin @FrancescoLocat8
1
8
20
2,284
Thrilled to announce that our paper is accepted at #NeurIPS 2025!! See you in San Diego! 🇺🇸
francesco ortu@francescortu10 Dec 2024
🚨🚨 Excited to share our latest paper, now on @arxiv! 🖼️ We studied how unified VLMs, trained to generate both text and images (e.g., @MetaAI's Chameleon), exchange information between modalities, comparing them to standard VLMs. Deep dive:👇
1
2
5
438
🚨🚨 Excited to share our latest paper, now on @arxiv! 🖼️ We studied how unified VLMs, trained to generate both text and images (e.g., @MetaAI's Chameleon), exchange information between modalities, comparing them to standard VLMs. Deep dive:👇
1
4
3
16
2,056
Lorenzo Basile retweeted
Diego@DiegoDoimo10 Dec 2024
I just landed in Vancouver to present @NeurIPSConf the findings of our new work! Few-shot learning and fine-tuning change the hidden layers of LLMs in a dramatically different way, even when they perform equally well on multiple-choice question-answering tasks. 🧵1/6
1
5
2
17
1,299
✨ Meet #ResiDual, a novel perspective on the alignment of multimodal latent spaces! Think of it as a spectral "panning for gold" along the residual stream. It improves text-image alignment by simply amplifying task-related directions! 🌌🔍 arxiv.org/abs/2411.00246 [1/6]
2
11
30
3,071
Load more