Search for: All records

We introduce AutoVER, an Autoregressive model for Visual Entity Recognition. Our model extends an autoregressive Multimodal Large Language Model by employing retrieval augmented constrained generation. It mitigates low performance on out-of-domain entities while excelling in queries that require visual reasoning. Our method learns to distinguish similar entities within a vast label space by contrastively training on hard negative pairs in parallel with a sequence-to-sequence objective without an external retriever. During inference, a list of retrieved candidate answers explicitly guides language generation by removing invalid decoding paths. The proposed method achieves significant improvements across different dataset splits in the recently proposed Oven-Wikibenchmark with accuracy on the Entity seen split rising from 32.7% to 61.5%. It demonstrates superior performance on the unseen and query splits by a substantial double-digit margin, while also preserving the ability to effectively transfer to other generic visual question answering benchmarks without further training. 

Abstract Stress‐based postseismic deformation modeling including afterslip and viscoelastic relaxation usually assumes the coseismic slip distribution and the associated stress perturbation as known. However, that assumption biases the postseismic modeling results by the assumptions that underlie the coseismic models. Importantly, this misses an opportunity to iteratively constrain the coseismic slip model with postseismic observations. We used a broad set of seismic and geodetic data to create multiple coseismic slip models that only differ in the down‐dip extent of the rupture plane and fit the coseismic observations for the July 29, Mw 8.2 Chignik earthquake equally well. We then evaluated the quality of those coseismic slip models based on how well each of them predicts postseismic GNSS displacements using a stress‐driven afterslip model. We find that coseismic slip models that generate afterslip too far down‐dip systematically fail to predict postseismic deformation. We find that the postseismic observations are best predicted by a narrower coseismic slip model that terminates abruptly at its deepest extent. The model predictions improve further if stress‐driven afterslip is combined with a superimposed viscoelastic relaxation response of a 50 km thick elastic lithosphere for the overriding plate and an elastic cold nose to the mantle wedge. 

The EIC Comprehensive Chromodynamics Experiment (ECCE) detector has been designed to address the full scope of the proposed Electron Ion Collider (EIC) physics program as presented by the National Academy of Science and provide a deeper understanding of the quark–gluon structure of matter. To accomplish this, the ECCE detector offers nearly acceptance and energy coverage along with excellent tracking and particle identification. The ECCE detector was designed to be built within the budget envelope set out by the EIC project while simultaneously managing cost and schedule risks. This detector concept has been selected to be the basis for the EIC project detector.