Now online (Neurocomputing), led by Nhut Truong, "Beyond topography: Topographic regularization improves robustness and reshapes representations in convolutional neural networks". Improvements found for image-noise, weight perturbations and adversarial attacks. www.sciencedirect.com/science/arti...

For those for whom tomorrow is always the 35th of May rather than 4/6, best wishes for a surreal day where, as Uncle Ringelhuth said, anything is possible.

Can we identify consistent attributions in DNNs trained to predict human authenticity judgments of AI-generated images? We find strong consistency within, but not across architectures. Relatedly, some DNNs seem to learn image-quality cues rather than authenticity-specific information.

arxiv cs.CV@arxiv-cs-cv.bsky.social · 4mo ago

Icaro Re Depaolini, Uri Hasson Non-identifiability of Explanations from Model Behavior in Deep Networks of Image Authenticity Judgments https://arxiv.org/abs/2604.07254

The NSF 2027 budget has noted that they will close out the Social, Behavioral, and Economic Science Program (SBE). This is not a good thing. nsf-gov-resources.nsf.gov/files/FY-202...

#BIDS has been extended to #EyeTracking data! There is now a standard for organizing and sharing eye tracking, covering gaze position, pupil size, meta data, messages, and more. Great news for #OpenScience! 🎉 Martin Szinte et al: www.biorxiv.org/content/10.6...

Why you should be very careful installing packages from Github: github.com/BerriAI/lite...

[Security]: CRITICAL: Malicious litellm_init.pth in litellm 1.82.8 — credential stealer · Issue #24512 · BerriAI/litellm

[Security]: CRITICAL: Malicious litellm_init.pth in litellm 1.82.8 PyPI package — credential stealer Summary The litellm==1.82.8 wheel package on PyPI contains a malicious .pth file (litellm_init.p...

github.com

מהדורה שניה, יוני 1980.

Olivia > Academics, check if your book is on the list of the Anthropic AI $1.5 billion settlement. Each author will get at least $3,000 per book. Claims must be made by March 30th. Repost to help authors whose works were illegally appropriated. secure.anthropiccopyrightsettlement.com/lookup?cultu...

Applications are open for the 2026/27 MSc in Cognitive Science at @cimecunitrento.bsky.social. First-call deadline for non-EU applicants: 𝗠𝗮𝗿𝗰𝗵 𝟰, 𝟮𝟬𝟮𝟲. Apply now: corsi.unitn.it/en/cognitive... #cimec_unitrento #neuroscience

How does the brain generate predictive models of own actions? We will soon open a **Postdoc position** to address this question in my lab. If you are interested, please write to moritz.wurm@unitn.it.

The results challenge the texture-bias hypothesis of Geirhos et al. (2019). This is one of those cases where a deep, careful review can add real value. arxiv.org/abs/2509.20234 🧠🤖 #MLSky

ImageNet-trained CNNs are not biased towards texture: Revisiting feature reliance through controlled suppression

The hypothesis that Convolutional Neural Networks (CNNs) are inherently texture-biased has shaped much of the discourse on feature use in deep learning. We revisit this hypothesis by examining limitat...

arxiv.org

@annabavaresco.bsky.social and @tlmnhut.bsky.social show: supervised pruning of a DNN’s feature space better aligns with human category representations, selects distinct subspaces for different categories, and more accurately predicts people’s preferences for GenAI images. doi.org/10.1145/3768...

Modeling Human Concepts with Subspaces in Deep Vision Models | ACM Transactions on Interactive Intelligent Systems

Improving the modeling of human representations of everyday semantic categories, such as animals or food, can lead to better alignment between AI systems and humans. Humans are thought to represent su...

doi.org

New preprint out together with @mheilbron.bsky.social We find that a stimulus' representational magnitude—the L2 norm of its DNN representation—predicts intrinsic memorability not just for images, but for words too. www.biorxiv.org/content/10.1...

Representational magnitude as a geometric signature of image and word memorability

What makes some stimuli more memorable than others? While memory varies across individuals, research shows that some items are intrinsically more memorable, a property quantifiable as “memorability”. ...

biorxiv.org

New lab preprint, led by @tlmnhut.bsky.social. We show that certain topographic CNNs offer computational advantages, including greater weight matrix robustness, better handling of OOD noisy data, and higher entropy of unit activation. arxiv.org/abs/2508.00043

Improved Robustness and Functional Localization in Topographic CNNs Through Weight Similarity

Topographic neural networks are computational models that can simulate the spatial and functional organization of the brain. Topographic constraints in neural networks can be implemented in multiple w...

arxiv.org

Nhut Truong, Uri Hasson: Improved Robustness and Functional Localization in Topographic CNNs Through Weight Similarity https://arxiv.org/abs/2508.00043 https://arxiv.org/pdf/2508.00043 https://arxiv.org/html/2508.00043

Does it seem like Spam is just mocking us at this point? "I hope your day is as positive as you are. We are very fortunate to associate with eminent people like you.... If you dont want to receive our email, reply as Terminate."

#Calcium_imaging in sleeping #bees: Antennal lobe neurons synchronise stronger during #sleep, likely due to reduced GABAergic coupling. #SNN simulations show reduced #odour processing, similar to human sleep: doi.org/10.1016/j.ne... Thanks to all collaborators at #CIMeC, e.g. @urihasson.bsky.social

Explore vision, motor control & predictive processing with a funded PhD @cimecunitrento.bsky.social Project: Predictive neural representations of dynamic events, supervisor Moritz Wurm @moritzwurm.bsky.social 🧠 Apply by June 5, 2025 🔗 Info: phd.unitn.it/drcimec #neuroscience #cognitivescience

Now online: We used two-photon calcium imaging to provide the first demonstration of how topological signatures of bees’ brain connectivity differ between sleep and wakefulness. Project guided by Albrecht Haase; first author: Sebastián Moguilner. www.sciencedirect.com/science/arti...

Neuronal correlates of sleep in honey bees

Honey bees Apis mellifera follow the day-night cycle for their foraging activity, entering rest periods during darkness. Despite considerable research…

sciencedirect.com

PhD opening. We've been working in recent years on human-oriented and explainable AI, studying interpretable DNN models of conceptual knowledge. Looking for a PhD student (4 years, Trento, Italy) to develop that direction. If interested, get in touch by email: my first name at my last name dot org

DNNs can predict human similarity judgments—but why? In a new #XAI study, we introduce Alignment Importance Scores (AIS), a method that improves AI-human alignment and generates heatmaps highlighting the image features that drive this alignment. link.springer.com/article/10.1...

Explaining Human Comparisons Using Alignment-Importance Heatmaps - Computational Brain & Behavior

We present a computational explainability approach for human comparison tasks, using Alignment Importance Score (AIS) heatmaps derived from deep-vision models. The AIS reflects a feature map’s unique ...

link.springer.com

Just received notice that our grant was terminated. 5R24AG066599-03 TheRiseRegistry.org

ICCV has an interesting policy that can be easily adopted by other societies.

#ICCV2025@iccv.bsky.social · last yr.

Check out the changes for #ICCV2025 🌶️ Full details: iccv.thecvf.com/Conferences/...

Department of Transportation orders all personnel to "identify and eliminate" every order, directive, rule, regulation, policy, notice, guidance document, funding arrangement, or program that even mentions climate change, diversity, or environmental justice www.transportation.gov/sites/dot.go...

As a former NSF program director, I find it difficult to put into words the extent to which the content and tone of this memo go against the grain of anything and everything that defines merit-based review at NSF.

Interesting analyses from Hanson et al., the strain on scientific publishing. Many more articles, more special issues (namely Frontiers, MDPI), but stagnation in number of PHDs granted. "in 2022 the article total was ∼47% higher than in 2016". direct.mit.edu/qss/article/...

New paper with Giuseppe Notaro presenting a method for quantitative measurement of sighting eye dominance link.springer.com/article/10.3...

Quantifying sighting dominance using on-display projections of monocular and binocular views - Behavior Research Methods

Sighting dominance is an important behavioral property which has been difficult to measure quantitatively with high precision. We developed a measurement method that is grounded in a two-camera model ...

link.springer.com

The neurobiology of language does not operate in a vat. An important perspective from the @thelablab.bsky.social and colleagues: "Language is widely distributed throughout the brain" www.nature.com/articles/s41...

1/n Some time ago my colleague, excellent cook, and friend Ivan told me: "Cacio e pepe is the recipe that I screw up more often. Let's make a project studying systematically the physics of that sauce". Prepare to get cheesy, I'm glad to share the Cacio e paper preprint: arxiv.org/abs/2501.00536

I'm at #NeurIPS2024 this week! Excited to present our work on perceptual alignment with Stephanie Fu on Thursday morning (poster #1302). Please reach out if you want to chat about synthetic data, alignment, data-centric AI, representation learning, or anything else!

Phillip Isola@phillipisola.bsky.social · 2y ago

First up, "When Does Perceptual Alignment Benefit Vision Representations?" (arxiv.org/abs/2410.10817) This paper is about distance between embeddings. It says: measure how humans perceive distance, then adjust a neural net to match. This improves transfer to lots of tasks (but not all tasks).