Searcharxiv⌕ Search

arXiv subjects

Nahal Mafi

Publications and source records attributed to Nahal Mafi.

2 recordsLinked to original sources

Does Pose Encoding Matter? Analyzing Geometric Representations for Discrete Motion Vocabulary Learning

Compressing continuous human motion into a finite vocabulary of discrete tokens is now the first stage of many motion understanding and generation pipelines. Vector quantization partitions a latent space into Voronoi cells around a finite codebook, but the geometry of that latent space is inherited from the input pose encoding, and it is unclear which encoding admits an efficient discretization. We study four encodings that place skeletal motion in different geometric spaces: Gaussian joint heatmaps on the pixel lattice, which we show to be a morphological dilation of the joint point set by a quadratic structuring function; root-relative joint positions in Euclidean space; local joint rotations on $\SO$ through a continuous 6D embedding; and a redundant full kinematic encoding. For each, we train VQ-VAE tokenizers whose codebooks lie on the unit hypersphere, varying codebook size, temporal stride, and code dimension. Beyond reconstruction, we measure codebook utilization and the effective vocabulary size, i.e., the number of equiprobable spherical Voronoi cells with the same entropy as the observed token distribution. Rotation and position encodings keep converting additional codewords into effective vocabulary, whereas the full kinematic encoding saturates at a few hundred effective cells and heatmaps at fewer than sixty. The smallest code dimension gives the highest utilization for three of the four encodings, and the effect of temporal stride depends on the encoding. Encodings of the same pose family can thus yield effective vocabularies differing by an order of magnitude, a difference that reconstruction loss alone does not reveal, which points to input geometry as a first-order design choice in motion tokenization.

cs.CV↗

Beyond the Interface: Redefining UX for Society-in-the-Loop AI Systems

Artificial intelligence systems increasingly operate in decision-critical environments where probabilistic outputs and Human-in-the-Loop (HITL) interactions reshape user engagement. Traditional user experience (UX) frameworks, designed for deterministic systems, fail to capture these evolving sociotechnical dynamics. This paper argues that in AI-enabled HITL systems, UX must transcend frontend usability to encompass backend performance, organizational workflows, and decision making structures. We employ a mixed-methods approach, combining an inductive social construction analysis of 269 stakeholder insights with the deployment of an operational HITL video anomaly detection system. Our findings reveal that stakeholders experience AI through multifaceted themes: risk, governance, and organizational capacity. Experimental results further demonstrate how detection behavior and alert routing directly calibrate human oversight and workload. Grounded in these results, we formalize a new evaluative framework centered on four sociotechnical metrics: Accuracy (FPR/FNR), Operational Latency (response time), Adaptation Time (deployment burden), and Trust (validated automation scales). This framework redefines UX as a multi-layered construct spanning infrastructure and governance, providing a rigorous foundation for evaluating AI systems embedded within complex real-world ecosystems.

cs.HC↗