Istituto Italiano di Tecnologia

About: Istituto Italiano di Tecnologia is a facility organization based out in Genoa, Italy. It is known for research contribution in the topics: Robot & Humanoid robot. The organization has 4561 authors who have published 14595 publications receiving 437558 citations. The organization is also known as: Italian Institute of Technology & IIT.

From Binary Cu2S to Ternary Cu–In–S and Quaternary Cu–In–Zn–S Nanocrystals with Tunable Composition via Partial Cation Exchange

Abstract: We present an approach for the synthesis of ternary copper indium sulfide (CIS) and quaternary copper indium zinc sulfide (CIZS) nanocrystals (NCs) by means of partial cation exchange with In3+ and Zn2+. The approach consists of a sequential three-step synthesis: first, binary Cu2S NCs were synthesized, followed by the homogeneous incorporation of In3+ by an in situ partial cation-exchange reaction, leading to CIS NCs. In the last step, a second partial exchange was performed where Zn2+ partially replaced the Cu+ and In3+ cations at the surface, creating a ZnS-rich shell with the preservation of the size and shape. By careful tuning reaction parameters (growth and exchange times as well as the initial Cu+:In3+:Zn2+ ratios), control over both the size and composition was achieved. This led to a broad tuning of photoluminescence of the final CIZS NCs, ranging from 880 to 1030 nm without altering the NCs size. Cytotoxicity tests confirmed the biocompatibility of the synthesized CIZS NCs, which opens up oppor...

Modality Distillation with Multiple Stream Networks for Action Recognition

Abstract: Diverse input data modalities can provide complementary cues for several tasks, usually leading to more robust algorithms and better performance. However, while a (training) dataset could be accurately designed to include a variety of sensory inputs, it is often the case that not all modalities are available in real life (testing) scenarios, where a model has to be deployed. This raises the challenge of how to learn robust representations leveraging multimodal data in the training stage, while considering limitations at test time, such as noisy or missing modalities. This paper presents a new approach for multimodal video action recognition, developed within the unified frameworks of distillation and privileged information, named generalized distillation. Particularly, we consider the case of learning representations from depth and RGB videos, while relying on RGB data only at test time. We propose a new approach to train an hallucination network that learns to distill depth features through multiplicative connections of spatiotemporal representations, leveraging soft labels and hard labels, as well as distance between feature maps. We report state-of-the-art results on video action classification on the largest multimodal dataset available for this task, the NTU RGB+D, as well as on the UWA3DII and Northwestern-UCLA.

Single-molecule pulling simulations can discern active from inactive enzyme inhibitors.

Abstract: Understanding ligand−protein recognition and interaction processes is of primary importance for structure-based drug design. Traditionally, several approaches combining docking and molecular dynamics (MD) simulations have been exploited to investigate the physicochemical properties of complexes of pharmaceutical interest. Even if the geometric properties of a modeled protein−ligand complex can be well predicted by computational methods, it is challenging to rank a series of analogues in a consistent fashion with biological data. In the unique β-hydroxyacyl-ACP dehydratase of Plasmodium falciparum (PfFabZ), the application of standard molecular docking and MD simulations was partially sufficient to shed light on the activity of previously discovered inhibitors. Complementing docking results with atomistic simulations in the steered molecular dynamics (SMD) framework, we devised an in silico approach to study molecular interactions and to compare the binding characteristics of ligand analogues. We hypothesi...