Curiously Effective Features For Image Quality Prediction
We find feature extractors constructed from random noise to be sufficient to learn a linear regression model whose quality predictions reach high correlations with human visual quality ratings, on par with a model with learned features.
EEG-Based Analysis of the Impact of Familiarity in the Perception of Deepfake Videos
Using EEG, we classify perceptual differences of familiar and unfamiliar people versus their face-swapped counterparts. Our results show that it is possible to discriminate fake videos from genuine ones when at least one face-swapped actor is...
An Efficient Multi-Link Channel Model for LiFi
The emergence of LiFi for indoor communications opens up new possibilities for wireless services in crowded multiuser scenarios. The deployment of LiFi in indoor scenarios is challenging due to the line-of-sight (LOS) blockage as well as...
Zero on Shape: A Generic 2D-3D Instance Similarity Metric learned from Synthetic Data
We present a network architecture which compares RGB images and untextured 3D models by the similarity of the represented shape. Our system is optimised for Zero-Shot retrieval, meaning it can recognise shapes never shown in training.
Enabling S-C-L-Band Systems with Standard C-Band Modulator and Coherent Receiver using Nonlinear Predistortion
To counteract the forthcoming capacity crunch in optical networks by increasing the throughput over already existing fiber infrastructures, innovative approaches are required. Multiband systems are an interesting approach to address this issue....
S-Band Transmission with Off-the-Shelf C-Band Components Enabled by Nonlinear Predistortion based on Coherent System Identification
Nonlinear Predistortion based on Coherent System Identification Text: In order to cope with the rapid traffic growth and as well reducing the cost-per-bit, reuse of the legacy optical fiber infrastructure is one of the main objectives for...
Clustered Federated Learning: Model-Agnostic Distributed Multi-Task Optimization under Privacy Constraints
Federated learning (FL) is the most widely adopted framework for collaborative training of deep learning models under privacy constraints. Albeit its popularity, it has been observed that FL yields suboptimal results if the clients’ data...
FantastIC4: A Hardware-Software Co-Design Approach for Efficiently Running 4bit-Compact Multilayer Perceptrons
With the growing demand for deploying Deep Learning models to the “edge”, it is paramount to develop techniques that allow to execute models within very tight and limited resource constraints. In this work we propose a software-hardware...
Explain and Improve: LRP-Inference Fine Tuning for Image Captioning Models
This paper analyzes the predictions of image captioning models with attention mechanisms beyond visualizing the attention itself. We compare the interpretability of attention heatmaps systematically against the explanations. We demonstrate that...
LiFi Positioning for Industry 4.0
Precise position information is considered as the main enabler for the implementation of smart manufacturing systems in Industry 4.0. In ...