Nicholas Sinigaglia | Università degli Studi di Padova (original) (raw)

Uploads

Papers by Nicholas Sinigaglia

Research paper thumbnail of Solving Cart pole and Racing car reinforcement learning tasks

In this report, I solve both cart-pole and racing car environments in using a DQN approach with e... more In this report, I solve both cart-pole and racing car environments in using a DQN approach with experience replay and fixed target. In the first case, the action value function is approximated by a fully connected NN, while in the second case, by a convolutional neural network (CNN). Due to the long needed computational time, in RL it is not wise to perform a "blind" gridsearch trying to find the best setup. For this reason, I focus on smart ways to define rewards and, for the racing car task, on useful and efficient pre-processing of input images.
This may be considered a process of "human learning", in which I study the two tasks focusing on the relevant features to be considered in order to make them easier to solve.

Research paper thumbnail of Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway)

International journal of building pathology and adaptation, Apr 28, 2022

Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties.... more Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties. Their physical deterioration is often the cumulative effect of different environmental hazards of variable intensity. Among these, fluctuations of temperature and relative humidity may cause nonrecoverable physical changes in building envelopes and artifacts made of hygroscopic materials, such as wood. Microclimatic fluctuations may be caused by several factors, including the presence of many visitors within the historical building. Within this framework, the current work is focused on detecting events taking place in two Norwegian stave churches, by identifying the fluctuations in temperature and relative humidity caused by the presence of people attending the public events. Design/methodology/approach-The identification of such fluctuations and, so, of the presence of people within the churches has been carried out through three different methods. The first is an unsupervised clustering algorithm here termed "density peak," the second is a supervised deep learning model based on a standard convolutional neural network (CNN) and the third is a novel ad hoc engineering feature approach "unexpected mixing ratio (UMR) peak." Findings-While the first two methods may have some instabilities (in terms of precision, recall and normal mutual information [NMI]), the last one shows a promising performance in the detection of microclimatic fluctuations induced by the presence of visitors. Originality/value-The novelty of this work stands in using both well-established and in-house ad hoc machine learning algorithms in the field of heritage science, proving that these smart approaches could be of extreme usefulness and could lead to quick data analyses, if used properly.

Research paper thumbnail of Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway)

Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway), 2022

Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties.... more Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties. Their physical deterioration is often the cumulative effect of different environmental hazards of variable intensity. Among these, fluctuations of temperature and relative humidity may cause nonrecoverable physical changes in building envelopes and artifacts made of hygroscopic materials, such as wood. Microclimatic fluctuations may be caused by several factors, including the presence of many visitors within the historical building. Within this framework, the current work is focused on detecting events taking place in two Norwegian stave churches, by identifying the fluctuations in temperature and relative humidity caused by the presence of people attending the public events. Design/methodology/approach-The identification of such fluctuations and, so, of the presence of people within the churches has been carried out through three different methods. The first is an unsupervised clustering algorithm here termed "density peak," the second is a supervised deep learning model based on a standard convolutional neural network (CNN) and the third is a novel ad hoc engineering feature approach "unexpected mixing ratio (UMR) peak." Findings-While the first two methods may have some instabilities (in terms of precision, recall and normal mutual information [NMI]), the last one shows a promising performance in the detection of microclimatic fluctuations induced by the presence of visitors. Originality/value-The novelty of this work stands in using both well-established and in-house ad hoc machine learning algorithms in the field of heritage science, proving that these smart approaches could be of extreme usefulness and could lead to quick data analyses, if used properly.

Research paper thumbnail of Analysis of real and generated food webs and study on pollution spreading

Foodwebs are nowadays an interesting and challenging subject of study for network analysis. In th... more Foodwebs are nowadays an interesting and challenging subject of study for network analysis. In this study real food webs are compared one to each other in terms of how authors collected data, in order to build a coherent dataset. Emergent patterns of these food webs are then quantified and three different generative models are used to try to reproduce them: cascade model, niche model and nested hierarchy model. A detailed comparison shows how niche model is the most suited of the three in accord with the data. Finally, an original clustering algorithm based on the idea of pollution spread is proposed and studied over food webs generated with the niche model. This algorithm needs no parameters, is suited for directed networks and provides information on which species may be screened to understand if the considered food web is polluted.

Research paper thumbnail of Low complexity CNN for time series classification

Time series analysis and classification are common tasks in many scientific fields. Nowadays, the... more Time series analysis and classification are common tasks in many scientific fields. Nowadays, the main tools used for their classification are the Convolutional Neural Networks: powerful machine learning algorithms with a high number of parameters very suitable for pattern recognition. The main focus of our work is to find a CNN architecture able to classify as well as possible simple univariate time series with the constraint of having at most 600 learnable parameters. In the following paper, after a brief introduction about the general framework of time series analysis, we present our self-designed model and its learning performance, comparing its classification ability with the results from a 2250-parameters CNN and from a 9968-parameters one showing how a few parameters CNN is a more suitable choice in the case of relatively small training set.

Research paper thumbnail of Approaches to Semantic Segmentation of Aerial Images

Image segmentation is a topic of paramount importance in our society, it finds applications from ... more Image segmentation is a topic of paramount importance in our society, it finds applications from computer vision to medical image analysis, robotic perception, video surveillance, and many others. Currently, there are various algorithms deployed for the semantic segmentation task and the most common are deep learning models such as convolutional neural networks (CNNs). In this report, we present, implement, and study three different algorithms to perform semantic segmentation of aerial images under the constraint of limited data and few classes. The first approach is a fully CNN (FCNN) designed by us taking inspiration from U-Net. The second approach is to adapt the Xception pretrained classifier using transfer learning and fine-tuning (XTFT). The third and last approach is a Random Forest Classifier (RF). The models are trained over the same dataset and in the same environment (same system specifics). Thanks to this, we provide a complete comparison of the three models, seeing how the best approach in our case is a FCNN with a contained number of parameters.

Drafts by Nicholas Sinigaglia

Research paper thumbnail of FPGA Implementation of a FIR Filter

The main goal of this report is to implement the FIR filtering procedure on a FPGA Xilinx Artix-7... more The main goal of this report is to implement the FIR filtering procedure on a FPGA Xilinx Artix-7 board, to study its drawbacks and to apply it over an example dataset.

Research paper thumbnail of Solving Cart pole and Racing car reinforcement learning tasks

In this report, I solve both cart-pole and racing car environments in using a DQN approach with e... more In this report, I solve both cart-pole and racing car environments in using a DQN approach with experience replay and fixed target. In the first case, the action value function is approximated by a fully connected NN, while in the second case, by a convolutional neural network (CNN). Due to the long needed computational time, in RL it is not wise to perform a "blind" gridsearch trying to find the best setup. For this reason, I focus on smart ways to define rewards and, for the racing car task, on useful and efficient pre-processing of input images.
This may be considered a process of "human learning", in which I study the two tasks focusing on the relevant features to be considered in order to make them easier to solve.

Research paper thumbnail of Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway)

International journal of building pathology and adaptation, Apr 28, 2022

Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties.... more Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties. Their physical deterioration is often the cumulative effect of different environmental hazards of variable intensity. Among these, fluctuations of temperature and relative humidity may cause nonrecoverable physical changes in building envelopes and artifacts made of hygroscopic materials, such as wood. Microclimatic fluctuations may be caused by several factors, including the presence of many visitors within the historical building. Within this framework, the current work is focused on detecting events taking place in two Norwegian stave churches, by identifying the fluctuations in temperature and relative humidity caused by the presence of people attending the public events. Design/methodology/approach-The identification of such fluctuations and, so, of the presence of people within the churches has been carried out through three different methods. The first is an unsupervised clustering algorithm here termed "density peak," the second is a supervised deep learning model based on a standard convolutional neural network (CNN) and the third is a novel ad hoc engineering feature approach "unexpected mixing ratio (UMR) peak." Findings-While the first two methods may have some instabilities (in terms of precision, recall and normal mutual information [NMI]), the last one shows a promising performance in the detection of microclimatic fluctuations induced by the presence of visitors. Originality/value-The novelty of this work stands in using both well-established and in-house ad hoc machine learning algorithms in the field of heritage science, proving that these smart approaches could be of extreme usefulness and could lead to quick data analyses, if used properly.

Research paper thumbnail of Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway)

Machine learning and engineering feature approaches to detect events perturbing the indoor microclimate in Ringebu and Heddal stave churches (Norway), 2022

Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties.... more Purpose-Climate-induced damage is a pressing problem for the preservation of cultural properties. Their physical deterioration is often the cumulative effect of different environmental hazards of variable intensity. Among these, fluctuations of temperature and relative humidity may cause nonrecoverable physical changes in building envelopes and artifacts made of hygroscopic materials, such as wood. Microclimatic fluctuations may be caused by several factors, including the presence of many visitors within the historical building. Within this framework, the current work is focused on detecting events taking place in two Norwegian stave churches, by identifying the fluctuations in temperature and relative humidity caused by the presence of people attending the public events. Design/methodology/approach-The identification of such fluctuations and, so, of the presence of people within the churches has been carried out through three different methods. The first is an unsupervised clustering algorithm here termed "density peak," the second is a supervised deep learning model based on a standard convolutional neural network (CNN) and the third is a novel ad hoc engineering feature approach "unexpected mixing ratio (UMR) peak." Findings-While the first two methods may have some instabilities (in terms of precision, recall and normal mutual information [NMI]), the last one shows a promising performance in the detection of microclimatic fluctuations induced by the presence of visitors. Originality/value-The novelty of this work stands in using both well-established and in-house ad hoc machine learning algorithms in the field of heritage science, proving that these smart approaches could be of extreme usefulness and could lead to quick data analyses, if used properly.

Research paper thumbnail of Analysis of real and generated food webs and study on pollution spreading

Foodwebs are nowadays an interesting and challenging subject of study for network analysis. In th... more Foodwebs are nowadays an interesting and challenging subject of study for network analysis. In this study real food webs are compared one to each other in terms of how authors collected data, in order to build a coherent dataset. Emergent patterns of these food webs are then quantified and three different generative models are used to try to reproduce them: cascade model, niche model and nested hierarchy model. A detailed comparison shows how niche model is the most suited of the three in accord with the data. Finally, an original clustering algorithm based on the idea of pollution spread is proposed and studied over food webs generated with the niche model. This algorithm needs no parameters, is suited for directed networks and provides information on which species may be screened to understand if the considered food web is polluted.

Research paper thumbnail of Low complexity CNN for time series classification

Time series analysis and classification are common tasks in many scientific fields. Nowadays, the... more Time series analysis and classification are common tasks in many scientific fields. Nowadays, the main tools used for their classification are the Convolutional Neural Networks: powerful machine learning algorithms with a high number of parameters very suitable for pattern recognition. The main focus of our work is to find a CNN architecture able to classify as well as possible simple univariate time series with the constraint of having at most 600 learnable parameters. In the following paper, after a brief introduction about the general framework of time series analysis, we present our self-designed model and its learning performance, comparing its classification ability with the results from a 2250-parameters CNN and from a 9968-parameters one showing how a few parameters CNN is a more suitable choice in the case of relatively small training set.

Research paper thumbnail of Approaches to Semantic Segmentation of Aerial Images

Image segmentation is a topic of paramount importance in our society, it finds applications from ... more Image segmentation is a topic of paramount importance in our society, it finds applications from computer vision to medical image analysis, robotic perception, video surveillance, and many others. Currently, there are various algorithms deployed for the semantic segmentation task and the most common are deep learning models such as convolutional neural networks (CNNs). In this report, we present, implement, and study three different algorithms to perform semantic segmentation of aerial images under the constraint of limited data and few classes. The first approach is a fully CNN (FCNN) designed by us taking inspiration from U-Net. The second approach is to adapt the Xception pretrained classifier using transfer learning and fine-tuning (XTFT). The third and last approach is a Random Forest Classifier (RF). The models are trained over the same dataset and in the same environment (same system specifics). Thanks to this, we provide a complete comparison of the three models, seeing how the best approach in our case is a FCNN with a contained number of parameters.

Research paper thumbnail of FPGA Implementation of a FIR Filter

The main goal of this report is to implement the FIR filtering procedure on a FPGA Xilinx Artix-7... more The main goal of this report is to implement the FIR filtering procedure on a FPGA Xilinx Artix-7 board, to study its drawbacks and to apply it over an example dataset.