Amin Rahimian

I've been working on a new tool, Refine, to make scholars more productive. If you're interested in being among the very first to try the beta, please read on.

Refine leverages the best current AI models to draw your attention to potential errors and clarity issues in research paper drafts.

1/

How information sharing in social media has slowly shifted from public posting to private channels. Strong empirical evidence from Facebook by the one and only Kiran Garimella. DETOX #ICWSM.

Assistant, Associate or Full Professor, AI & Society The Department of AI and Society (AIS) at the University at Buffalo (UB) invites candidates to apply for multiple positions as Assistant Professor, Associate Professor, or Full Professor. The new AIS ...

UB's new Department of AI and Society is hiring faculty across ranks (Assistant, Associate, Full Professor). We’re looking for transdisciplinary scholars interested in building AI by society, for society. Start dates begin Fall 2025.

More info: www.ubjobs.buffalo.edu/postings/57734

Great news! congrats:)

Nice new paper in PNAS providing further evidence that random/long ties help social contagions — even many that would be labeled "complex contagions"
www.pnas.org/doi/10.1073/... @davidlazer.bsky.social

Screenshot of the text of the linked blogpost 1/4

Screenshot of the text of the linked blogpost 2/4

Screenshot of the text of the linked blogpost 3/4

Screenshot of the text of the linked blogpost 4/4

~~ making sense of academic statistics ~~

i wrote about the confusing relationship between statistics and data analysis, and also about how statistics relates to science

#statistics #rstats #datascience

www.alexpghayes.com/post/making-...

ACM EAAMO, which is coming to Pitt this Fall, has two events for students: a doctoral consortium and a poster session, both of which are due July 25th
- poster session conference.eaamo.org/cfp/call_for...
- doctoral consortium
conference.eaamo.org/cfp/call_for...

ACM EAAMO, which is coming to Pitt this Fall, has two events for students: a doctoral consortium and a poster session, both of which are due July 25th
- poster session conference.eaamo.org/cfp/call_for...
- doctoral consortium
conference.eaamo.org/cfp/call_for...

Who Sees Who? In the comedy Home Alone, a burglar posing as a police officer walks door to door during the holidays to find out which families will be traveling, leaving their empty homes easy prey for a break-in. ...

- differentially private distributed estimation & learning arxiv.org/abs/2306.15865 IISE transactions, featured: content.presspage.com/uploads/2602...
- differentially private distributed inference arxiv.org/abs/2402.08156
- privacy-preserving sequential learning arxiv.org/abs/2502.19525 (FORC'25)

Who Sees Who? In the comedy Home Alone, a burglar posing as a police officer walks door to door during the holidays to find out which families will be traveling, leaving their empty homes easy prey for a break-in. ...

nice piece by Pitt Swanson School of Engineering, showcasing three recent privacy works with @papachristoumarios.bsky.social and @yuxin-pitt.bsky.social - news.engineering.pitt.edu/who-sees-who/

🚨Free data alert!! 🚨 Please share.

Large new dataset of Amazon product reviews, including full text and photos and product characteristics, with individual *reviews labeled as fake reviews*.

I believe this is the first publicly available data of this kind.

github.com/bretthollenb...

City-Defined Neighborhood Boundaries in the United States - Scientific Data Scientific Data - City-Defined Neighborhood Boundaries in the United States

"City-Defined Neighborhood Boundaries in the United States"
We provide a new dataset of city-defined neighborhoods for 206 of the largest cities in the United States, covering more than 77 million people. 1/

www.nature.com/articles/s41...

Postdoctoral Position | Center for Adaptive Rationality

We are hiring, @arc-mpib.bsky.social a postdoc for a project to investigate why citizens feel alienated from liberal democracy and how a shared sense of reality can be restored.
Work with @lfoswaldo.bsky.social @anaskozyreva.bsky.social, Ralph Hertwig and me:
www.mpib-berlin.mpg.de/2084802/2025...

Cognitive representations of social networks in isolated villages - Nature Human Behaviour Feltham et al. develop a sampling strategy to evaluate social network cognition across 82 Honduran villages, systematically mapping the underlying village networks.

New by @ericfeltham.bsky.social, Laura Forastiere, and @nachristakis.bsky.social: an extraordinarily ambitious effort to scale up and bring Krackhardt's classic work on cognitive social structures (CSSs) into the 21st century. Super excited to see it in print. www.nature.com/articles/s41...

People not only form social networks, they construct mental maps of them. People think about the ties between other people, including ties among individuals to whom they are not themselves directly connected. These “cognitive social networks” have rarely been studied. 1/

in a sequential social learning scenario, a sequence of agents make binary decisions on the basis of their private signals and information available to them from past choices. information cascade in this scenario refers to discounting of new private information in favor of past public actions

@yuxin-pitt.bsky.social presenting shortly in responsiblecomputing.org/forc-2025/ (extended abstract: drops.dagstuhl.de/entities/vol... - working paper: arxiv.org/abs/2502.19525) also last month at the Network Science in Economics conference, both in beautiful Stanford bsky.app/profile/yuxi...

smooth randomized response guarantees learning at log(n) rate that is much faster than the sqrt(log n) rate in the non-private case & achieves finite expected times to the first correct action and last incorrect action, both of which are infinite in the non-private case + same eps can minimize both

a simple randomized response that flips actions with constant probability will prevents any cascade and asymptotic learning, but we can use a smooth version that hides signal locations within a range; this way, actions can be flipped with a probability that decays smoothly outside the range

for Gaussian signals, agent n switches its action from 0 to 1 when its private signal exceeds a threshold t(l_n) as a function of the log-likelihood ratio of public belief. In the non-private case, l_n grows as sqrt(log n), and agents eventually (but very slowly) take the correction almost surely

threshold k for a cascade (the number where having k signals of one type more than the other, makes following agents ignore their private signal and initiate a cascade) is fixed over intervals of eps. At interval boundaries, decreasing eps increases k by one and correct cascade probability jumps up

for binary signals, eps-DP is achieved using a randomized response mechanism that flips actions with probability 1/(1+e^eps) until a cascade occurs. The probability of correct cascades does not increase compared to the non-private baseline, but it does not vary monotonically with eps either

can we improve sequential learning by limiting information leakage to reduce the stickiness of information traps? randomizing actions limits what can be learned about each individual’s signal from their actions a la differential privacy + has potential to improve the collective learning outcome

the sqrt(log n) rate is due to Hann-Caruthers, Martynov and Tamuz (2018) www.sciencedirect.com/science/arti... who also show for Gaussian signals it implies infinite expected time to the first correct action (& more generally to the last incorrect action onlinelibrary.wiley.com/doi/epdf/10....)

with continuous signals & unbounded likelihood ratios (eg Gaussians) learning occurs b/c agents eventually receive strong enough signals to pull them out of information traps, but can be very slow - asymptotic growth rate of the log-likelihood ratio of public belief is sqrt(log n) << n

with discrete signals information cascade prevents learning (Bikhchandani, Hirshleifer & Welch 1992 www.journals.uchicago.edu/doi/10.1086/...) b/c agent n rationally ignores its private signal and follows the herd majority, which is false with a non-zero probability (not decreasing in n)

information cascade disrupts information aggregation and prevents efficient social learning, despite there being enough information in n independent signals to determine the common binary state with high accuracy if the agents had access to each other’s private signals www.aeaweb.org/articles?id=...

in a sequential social learning scenario, a sequence of agents make binary decisions on the basis of their private signals and information available to them from past choices. information cascade in this scenario refers to discounting of new private information in favor of past public actions

CoCoMARL 2025 Call for Papers

🥥🥥 Hoping to submit to CoCoMARL @rl-conference.bsky.social but need more time ⌛😱? Well you're in luck - we're extending our deadline *1 week* to June 6th!

See our call for papers and the link to submit below! 🥥🥥

sites.google.com/view/cocomar...

Conformal-DP: Differential Privacy on Riemannian Manifolds via Conformal Transformation Peilin He, Liou Tang, M. Amin Rahimian, James Joshi http://arxiv.org/abs/2504.20941 Differential Privacy (DP) has been established as a safeguard for private data sharing by adding perturbations to information release. Prior research on DP has extended beyond data in the flat Euclidean space and addressed data on curved manifolds, e.g., diffusion tensor MRI, social networks, or organ shape analysis, by adding perturbations along geodesic distances. However, existing manifold-aware DP methods rely on the assumption that samples are uniformly distributed across the manifold. In reality, data densities vary, leading to a biased noise imbalance across manifold regions, weakening the privacy-utility trade-offs. To address this gap, we propose a novel mechanism: Conformal-DP, utilizing conformal transformations on the Riemannian manifold to equalize local sample density and to redefine geodesic distances accordingly while preserving the intrinsic geometry of the manifold. Our theoretical analysis yields two main results. First, we prove that the conformal factor computed from local kernel-density estimates is explicitly data-density-aware; Second, under the conformal metric, the mechanism satisfies $ \varepsilon $-differential privacy on any complete Riemannian manifold and admits a closed-form upper bound on the expected geodesic error that depends only on the maximal density ratio, not on global curvatureof the manifold. Our experimental results validate that the mechanism achieves high utility while providing the $ \varepsilon $-DP guarantee for both homogeneous and especially heterogeneous manifold data.

Conformal-DP: Differential Privacy on Riemannian Manifolds via Conformal Transformation

Peilin He, Liou Tang, M. Amin Rahimian, James Joshi

http://arxiv.org/abs/2504.20941