Institution profile

Universitat Politècnica de Catalunya

Academic institutioneurope · es
Official website
Research library367linked papers
Opportunities0open roles
Selected work

Representative Papers

Frequency Selection for the Diagnostic Characterization of Human Brain Tumours

Jul 22, 2009International Conference of the Catalan Association for Artificial Intelligence

Conventional magnetic resonance spectroscopy (MRS) full-spectrum analysis for noninvasive brain tumor diagnosis suffers from high spectral redundancy and weak discriminability, limiting diagnostic accuracy and interpretability. Method: This paper proposes a task-oriented adaptive frequency-point selection method that jointly optimizes with nonlinear classifiers (e.g., SVM or ANN) to automatically identify the most discriminative spectral regions from high-dimensional metabolic profiles. Unlike traditional fixed-bandwidth or full-spectrum modeling approaches, our method explicitly balances feature interpretability and classification performance. Contribution/Results: Evaluated on an international multicenter brain tumor MRS dataset, the proposed method significantly reduces redundant frequency-point interference and achieves superior classification accuracy compared to state-of-the-art methods. It delivers clinically actionable, interpretable, and highly accurate auxiliary diagnostic support for neuro-oncology.

10 citations1 influentialRead paper

Opportunities and Challenges for Virtual Reality Streaming over Millimeter-Wave: An Experimental Analysis

Oct 05, 2022International Conference on Network of the Future

This study addresses transmission instability of millimeter-wave (mmWave) VR streaming under mobile and dynamic occlusion conditions. We develop the first experimental 802.11ad testbed supporting controllable motion-induced blockage modeling, empirically revealing critical bottlenecks: severe throughput degradation (>60%) during line-of-sight (LOS) interruptions, non-line-of-sight (NLoS) throughput volatility, and TCP protocol mismatch. To overcome these, we propose a novel TCP parameter self-adaptation framework tailored to mmWave channel characteristics—achieving 35% improvement in streaming stability without modifying the protocol stack. Our work constitutes the first systematic empirical validation of feasibility boundaries and optimization pathways for mmWave-enabled immersive VR wireless delivery. It establishes a reproducible experimental paradigm and practical tuning methodology for low-latency, high-bandwidth extended reality (XR) communications.

8 citationsRead paper

Work-Efficient Parallel Non-Maximum Suppression Kernels

Aug 21, 2020Computer/law journal

To address the real-time non-maximum suppression (NMS) bottleneck caused by thousands of overlapping candidate bounding boxes in embedded GPU-based object detection (e.g., NVIDIA Tegra X1/X2), this paper proposes a workload-balanced parallel greedy NMS algorithm. The method introduces a lightweight, dynamically load-partitioned CUDA kernel design—first of its kind—supporting variants such as FeatureNMS and Soft-NMS, thereby balancing accuracy and scalability. It achieves high optimization through deep architectural co-design with embedded GPU constraints, including memory bandwidth limitations and compute unit characteristics, enabling efficient thread scheduling and memory access patterns. On Tegra X1/X2, processing 1,024 candidate boxes requires only ~1 ms—14× to 40× faster than current state-of-the-art CNN-based learned NMS methods—significantly enhancing end-side real-time detection throughput and efficiency.

5 citations1 influentialRead paper

Generating Realistic Synthetic Head Rotation Data for Extended Reality using Deep Learning

Oct 10, 2022IXR@MM

To address the scarcity and high acquisition cost of real-world head-motion time-series data for XR systems, this work pioneers the adaptation of TimeGAN to head-rotation sequence modeling. We propose a conditional multivariate time-series generation framework that jointly models angular velocity and Euler angles, integrating LSTM-based generators and discriminators to ensure both dynamic consistency and statistical fidelity. Evaluated on real head-motion datasets, our synthesized data achieves a 37% reduction in Fréchet Inception Distance (FID), yields a 29% error reduction in downstream head-motion prediction models, and attains 92% perceptual realism as validated by expert blind evaluation. This work overcomes the critical bottleneck of head-motion data scarcity, substantially enhancing the generalization capability of prediction models. It establishes a novel paradigm for generating high-fidelity synthetic head-motion data, enabling improved real-time rendering and interaction in XR applications.

3 citationsRead paper

Unsupervised Cognition

Sep 27, 2024arXiv.org

Existing unsupervised learning methods exhibit significant limitations in cognitive plausibility and classification performance under few-shot or incomplete data regimes, particularly lacking human-like generalization and robustness. This paper introduces the first computationally grounded, cognition-inspired unsupervised learning paradigm grounded in primitive cognitive operations: a representation-centric framework that constructs input-agnostic, distributed, hierarchical representations, augmented by an unsupervised decision mapping mechanism enabling end-to-end classification. Departing from conventional clustering-based paradigms, our approach achieves state-of-the-art performance on unsupervised image classification, few-shot classification, and cancer subtype identification. Crucially, it surpasses supervised baselines on cognitive behavioral metrics—including cross-domain generalization, noise robustness, and decision consistency—demonstrating, for the first time, unsupervised models that emulate human-like cognitive reasoning at the behavioral level.

2 citations1 influentialRead paper
Recent publications

Latest Papers