arXiv subjects

Chandon Hamel

Publications and source records attributed to Chandon Hamel.

3 recordsLinked to original sources

arXiv2026

FlowMPC: Improving Flow Matching policies with World Models

Flow Matching (FM) is a powerful approach for behavior cloning in multimodal action spaces [Jiang et al., 2025], but because it is not trained to directly maximize expected return, there is still room to improve how FM policies act at test time. This work investigates whether a learned world model can improve FM policies by enabling Model Predictive Path Integral (MPPI) planning over candidate action sequences proposed by the policy. Building on TD-MPC2 [Hansen et al., 2024], I introduce FlowMPC, a framework that combines an imitation-learned FM policy with a learned world model for test-time planning in ManiSkill manipulation tasks [Tao et al., 2025]. Across PickCube and PickSingleYCB, adding the world model improved performance over the FM policy alone, with especially clear gains in end-of-episode success. These results suggest that world-model-based planning can effectively complement flow-based imitation policies without modifying the FM training objective.

cs.LG↗

arXiv2025

Analysis of Bias in Deep Learning Facial Beauty Regressors

Bias can be introduced to AI systems even from seemingly balanced sources, and AI facial beauty prediction is subject to ethnicity-based bias. This work sounds warnings about AI's role in shaping aesthetic norms while providing potential pathways toward equitable beauty technologies through comparative analysis of models trained on SCUT-FBP5500 and MEBeauty datasets. Employing rigorous statistical validation (Kruskal-Wallis H-tests, post hoc Dunn analyses). It is demonstrated that both models exhibit significant prediction disparities across ethnic groups $(p < 0.001)$, even when evaluated on the balanced FairFace dataset. Cross-dataset validation shows algorithmic amplification of societal beauty biases rather than mitigation based on prediction and error parity. The findings underscore the inadequacy of current AI beauty prediction approaches, with only 4.8-9.5\% of inter-group comparisons satisfying distributional parity criteria. Mitigation strategies are proposed and discussed in detail.

cs.CV↗

arXiv2025

GANji: A Framework for Introductory AI Image Generation

The comparative study of generative models often requires significant computational resources, creating a barrier for researchers and practitioners. This paper introduces GANji, a lightweight framework for benchmarking foundational AI image generation techniques using a dataset of 10,314 Japanese Kanji characters. It systematically compares the performance of a Variational Autoencoder (VAE), a Generative Adversarial Network (GAN), and a Denoising Diffusion Probabilistic Model (DDPM). The results demonstrate that while the DDPM achieves the highest image fidelity, with a Fr\'echet Inception Distance (FID) score of 26.2, its sampling time is over 2,000 times slower than the other models. The GANji framework is an effective and accessible tool for revealing the fundamental trade-offs between model architecture, computational cost, and visual quality, making it ideal for both educational and research purposes.

cs.CV↗