arXiv subjects

Alexander Hackett

Publications and source records attributed to Alexander Hackett.

3 recordsLinked to original sources

arXiv2026

From Recovery to Drop-off: How Action Post-training Reduces a VLM's Late-Layer Depth Decodability

How much of a vision-language model's (VLM) spatial understanding remains after the action post-training process of building a vision-language-action model (VLA)? We probe depth perception, a primitive of spatiogeometric understanding, from every decoder layer of a weight-matched open-source base VLM/VLA pair: Molmo2-ER and MolmoAct2-LIBERO. First, the VLA decodes depth worse at every layer, a persistent gap we call the floor. Second, the degradation is not uniform: while the base VLM's depth decodability improves through its final layers, the VLA's collapses, an additional late-layer drop we call the cliff. We causally localize the cliff to late-layer MLP interference: ablating the late-layer MLP writes recovers the majority of the terminal decodability cliff, while matched attention ablations and the same intervention in the weight-matched base VLM produce no comparable recovery. A module-level decomposition explains this dissociation: the base VLM carries depth most accessibly in accumulated MLP writes, whereas action post-training collapses depth decodability in the late accumulated writes.

cs.CV↗

arXiv2026

Pretraining Objective Matters in Extreme Low-Data FGVC: A Backbone-Controlled Study

Extreme low-data fine-grained classification is common in expert domains where labeling is expensive, yet practitioners still need principled guidance for selecting pretrained encoders. We study emerald inclusion grading with a custom dataset of labeled images across three classes and ask: under matched backbone capacity, how does pretraining objective affect downstream representation quality? We compare four frozen ViT-B/16 encoders trained with supervised classification, contrastive learning (SigLIP2), masked reconstruction (MAE), and self-distillation (DINOv3), and evaluate them with leave-one-out cross-validation using linear and nonlinear probes. To control statistical noise in the low-N regime, we use permutation testing (N=1000) on macro one-vs-rest AUC. Supervised and contrastive encoders provide the strongest linear separability (logistic AUC: 0.768 and 0.735; SVM AUC: 0.739 and 0.697), while MAE improves under nonlinear probes (XGBoost AUC: 0.713). We find that DINOv3 underperforms across probe families in this domain. These results support a practical recommendation for extreme low-data FGVC: prioritize margin-enforcing pretraining objectives when data scarcity restricts probing to linear decision rules, and consider reconstruction-style encoders when nonlinear classifiers are feasible given dataset constraints.

cs.CV↗

arXiv2021

Ionizing photon production of Population III stars: effects of rotation, convection, and initial mass function

The first stars are thought to be one of the dominant sources of hydrogen reionization in the early Universe, with their high luminosities and surface temperatures expected to drive high ionizing photon production rates. In this work, we take our Geneva stellar evolution models of zero-metallicity stars and predict their production rates of photons capable to ionize H, He I and He II, based on a blackbody approximation. We present analytical fits in the range 1.7-500 solar masses. We then explore the impact of stellar initial mass, rotation, and convective overshooting for individual stars. We have found that ionizing photon production rates increase with increasing initial mass. For the rotational velocities considered we see changes of up to 25% to ionizing photons produced. This varies with initial mass and ionizing photon species and reflects changes to surface properties due to rotation. We have also found that higher convective overshooting increases ionizing photon production by approximately 20% for the change in overshooting considered here. For stellar populations, we explore how the production of ionizing photons varies as a function of the initial mass function (IMF) slope, and minimum and maximum initial masses. For a fixed population mass we have found changes of the order of 20-30% through varying the nature of the IMF. This work presents ionizing photon production predictions for the most up to date Geneva stellar evolution models of Population III stars, and provides insight into how key evolutionary parameters impact the contribution of the first stars to reionization.

astro-ph.SR↗