arXiv subjects

Alex Rojas

Publications and source records attributed to Alex Rojas.

3 recordsLinked to original sources

arXiv2026

Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning

Standard autoregressive video generation algorithms based on Diffusion and Flow Matching rely on rigid training objectives and static sampling schedules, limiting inference procedures from adapting to the data. We introduce Equilibrium Forcing (EqF), a simplified framework for video denoising generative models without noise level conditioning. EqF pioneers modular training- and inference-time designs for noise-unconditional generation that decouple learning the denoising field from sampling. This flexibility allows for inference-time algorithms that operate in a closed loop by adapting to feedback from the sample, improving video quality and consistency on challenging autoregressive video generation benchmarks. Extensive analysis elucidates exactly how removing the noise level conditioning enables EqF's data-dependent inference properties to surpass the performance of standard noise level-conditional denoising video methods.

cs.CV↗

arXiv2024

Understanding the Role of Functional Diversity in Weight-Ensembling with Ingredient Selection and Multidimensional Scaling

Weight-ensembles are formed when the parameters of multiple neural networks are directly averaged into a single model. They have demonstrated generalization capability in-distribution (ID) and out-of-distribution (OOD) which is not completely understood, though they are thought to successfully exploit functional diversity allotted by each distinct model. Given a collection of models, it is also unclear which combination leads to the optimal weight-ensemble; the SOTA is a linear-time ``greedy" method. We introduce two novel weight-ensembling approaches to study the link between performance dynamics and the nature of how each method decides to use apply the functionally diverse components, akin to diversity-encouragement in the prediction-ensemble literature. We develop a visualization tool to explain how each algorithm explores various domains defined via pairwise-distances to further investigate selection and algorithms' convergence. Empirical analyses shed perspectives which reinforce how high-diversity enhances weight-ensembling while qualifying the extent to which diversity alone improves accuracy. We also demonstrate that sampling positionally distinct models can contribute just as meaningfully to improvements in a weight-ensemble.

cs.LG↗

arXiv2014

Galaxy And Mass Assembly (GAMA): the wavelength-dependent sizes and profiles of galaxies revealed by MegaMorph

We investigate the relationship between colour and structure within galaxies using a large, volume-limited sample of bright, low-redshift galaxies with optical to near-infrared imaging from the GAMA survey. We fit single-component, wavelength-dependent, elliptical Sersic models to all passbands simultaneously, using software developed by the MegaMorph project. Dividing our sample by Sersic index and colour, the recovered wavelength variations in effective radius (R_e) and Sersic index (n) reveal the internal structure, and hence formation history, of different types of galaxies. All these trends depend on n; some have an additional dependence on galaxy colour. Late-type galaxies (n_r < 2.5) show a dramatic increase in Sersic index with wavelength. This might be a result of their two-component (bulge-disk) nature, though stellar population gradients within each component and dust attenuation are likely to play a role. All galaxies show a substantial decrease in R_e with wavelength. This is strongest for early-types (n_r > 2.5), even though they maintain constant n with wavelength, revealing that ellipticals are a superimposition of different stellar populations associated with multiple collapse and merging events. Processes leading to structures with larger R_e must be associated with lower metallicity or younger stellar populations. This appears to rule out the formation of young cores through dissipative gas accretion as an important mechanism in the recent lives of luminous elliptical galaxies.

astro-ph.GA↗