arXiv subjects

Cristian Bosch

Publications and source records attributed to Cristian Bosch.

4 recordsLinked to original sources

arXiv2026

Uncertainty Makes It Stable: Curiosity-Driven Quantized Mixture-of-Experts

Deploying deep neural networks on resource-constrained devices faces two critical challenges: maintaining accuracy under aggressive quantization while ensuring predictable inference latency. We present a curiosity-driven quantized Mixture-of-Experts framework that addresses both through Bayesian epistemic uncertainty-based routing across heterogeneous experts (BitNet ternary, 1-16 bit BitLinear, post-training quantization). Evaluated on audio classification benchmarks (ESC-50, Quinn, UrbanSound8K), our 4-bit quantization maintains 99.9 percent of full-precision F1 (0.858 vs 0.859) with 4x compression and 31 percent energy savings versus 8-bit, while both achieve statistical parity with full precision (p > 0.05). Crucially, curiosity-driven routing simultaneously improves accuracy and stability: on Quinn, F1 increases from 0.802 to 0.809 while cross-fold variance drops by 85 percent (p < 0.001, Levene's test), with reductions of 50 to 94 percent across datasets. The routing is self-organizing, with the high-precision 8-bit expert automatically receiving the most uncertain samples (20 percent lower confidence, p < 0.001), while lightweight experts handle easier inputs. Datasets with already low baseline variance show no artificial stability gain, confirming the mechanism targets genuine epistemic uncertainty rather than overfitting routing decisions. At 1.2M parameters, the framework provides interpretable, precision-aware routing suitable for safety-sensitive edge deployments where both accuracy and predictability are critical.

cs.LG↗

arXiv2025

Embedding-Aware Quantum-Classical SVMs for Scalable Quantum Machine Learning

Quantum Support Vector Machines face scalability challenges due to high-dimensional quantum states and hardware limitations. We propose an embedding-aware quantum-classical pipeline combining class-balanced k-means distillation with pretrained Vision Transformer embeddings. Our key finding: ViT embeddings uniquely enable quantum advantage, achieving up to 8.02% accuracy improvements over classical SVMs on Fashion-MNIST and 4.42% on MNIST, while CNN features show performance degradation. Using 16-qubit tensor network simulation via cuTensorNet, we provide the first systematic evidence that quantum kernel advantage depends critically on embedding choice, revealing fundamental synergy between transformer attention and quantum feature spaces. This provides a practical pathway for scalable quantum machine learning that leverages modern neural architectures.

quant-ph↗

arXiv2016

A composite tale of two right-handed neutrinos

In this work, we develop a model for Higgs-like composites based on two generations of right handed neutrinos which condense. We analyze the Spontaneous Symmetry Breaking of the theory with two explicit breakings, setting the different scales of the model and obtaining massive bosons as a result. Finally, we calculate the gravitational wave imprint left by the phase transition associated to the symmetry breaking of a generic potential dictated by the symmetries of the composites.

hep-ph↗

arXiv2012

Prospects for constraining the shape of non-Gaussianity with the scale-dependent bias

We consider whether the non-Gaussian scale-dependent halo bias can be used not only to constrain the local form of non-Gaussianity but also to distinguish among different shapes. In particular, we ask whether it can constrain the behavior of the primordial three-point function in the squeezed limit where one of the momenta is much smaller than the other two. This is potentially interesting since the observation of a three-point function with a squeezed limit that does not go like the local nor equilateral templates would be a signal of non-trivial dynamics during inflation. To this end we use the quasi-single field inflation model of Chen and Wang as a representative two-parameter model, where one parameter governs the amplitude of non-Gaussianity and the other the shape. We also perform a model-independent analysis by parametrizing the scale-dependent bias as a power-law on large scales, where the power is to be constrained from observations. We find that proposed large-scale structure surveys (with characteristics similar to the dark energy task force stage IV surveys) have the potential to distinguish among the squeezed limit behavior of different bispectrum shapes for a wide range of fiducial model parameters. Thus the halo bias can help discriminate between different models of inflation.

astro-ph.CO↗