Importantly, we have demonstrated the first clear, positive result of emergent communication under competition in modern machine learning using standard RL. We have expanded on conditions for competitively learning to communicate and presented three…

We proposed a multi-task learning framework to solve two challenging and fundamental wireless signal recognition tasks - modulation and signal classification. We leveraged the relation between the two tasks in allowing the MTL to learn the shared re…

In this paper, we proposed a new network, D-net, and a new structure, Mutual Non-local Link (MNL), for estimation of transformation between CT volumes.

In conclusion, we presented AirWare, a technology that fuses the output of an embedded smart-phone microphone and proximity sensor to recognize a gesture set of 21 in-air hand-gestures with 50.47% average true positive rate per gesture. While we sho…

We developed a novel meaning representation and simulation-based annotation interface, enabling the collection of process-level annotations of experimental procedures, as well as two parsers (pipeline and joint modelling) trained on this data. Our d…

We propose a novel RGB-D SOD framework called RD3D, which is based on 3D CNNs and conducts cross-modal feature fusion in a progressive manner. RD3D first utilizes 3D convolutions for pre-fusion between RGB and depth, and then conduct explicit fusion…

In this paper, we proposed a Gaussian-based grasp representation to highlight the maximum grasp quality at the center position. Based on Gaussian-based grasp representation, a lightweight generative architecture with a receptive field block module a…

One of the latest challenges for both the clinical and the AI community has been applying deep learning in diagnosing COVID from CXRs. Recent works suggested the possibility of successfully tackling this problem, despite the currently small quantity…

Ida Mele is currently a researcher at IASI-CNR, Rome (Italy). She got her Ph.D. in Computer Engineering from Sapienza University of Rome. She has co-authored papers in peer-reviewed international conferences and top-tier journals. She has also serve…

In this paper, we propose a Trigger-Sense Memory Flow Framework (TriMF) for joint entity and relation extraction. We use the memory to boost the task-related information in a sentence through Multi-level Memory Flow Attention module. This module can…

We have shown that it is possible to distill the knowledge of not just a pre-trained model but also heuristically designed ISP, to improve the performance of a classifier for RAW images. We also show improvement for short-exposure RAW, distilling th…