Dumitru Iulian Nastac

Dr. Dumitru Iulian Nastac

orcid
avatar
Co-authors
avatar
Following
avatar
Followers

Research Focus & Metrics

Explore full metrics on Scilit

Academic Career Length

25Years (2002-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 September 2018
Verified
Conference Paper
Closed access
Alex AwadaAlex Awada
Alex Awada
IMIrina Mocanu
Irina Mocanu
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
DBDan Benta
Dan Benta
SRSerban Radu
Serban Radu
Adaptive user interfaces can be more effective in increasing accessibility of the interaction between people and systems, especially for elderly or people with different impairments such as cognitive impairments in people with dementia. There are two possible types of adaptation for interfaces: navigation and content-based adaptation. Both types of adaptation must be made based on the context (devices that are used) and on the user profile. This paper presents a method for interface adaptation using user classification. Users are classified based on their interactions with the interface.
1 January 2014 in International Journal of Process Management and Benchmarking
Verified
Journal Article
Closed access
EDEmilian Dobrescu
Emilian Dobrescu
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
EPElena Pelinescu
Elena Pelinescu
Our purpose is to verify the predictive performances of the artificial neural networks (ANNs) under volatile statistics and possibly incomplete information. Daily forecasts of exchange rate using exclusively primary available information for an emergent economy (such as the Romanian one) could be a proper experimental ground with such a goal. The present paper extends the previous authors' research (Dobrescu et al., 2006; Nastac et al., 2007) on the same issue to improve the accuracy of exchange rate forecasting by using a set of neural predictors in cascade, instead of a single one. The results show that the presented model, despite its proved advantages, could be further improved in order to avoid the translation into residuals of the high serial correlation present in the primary database.
1 October 2017
Verified
Conference Paper
Closed access
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
FIFlorentin Alexandru Iftimie
Florentin Alexandru Iftimie
OAOctavian Arsene
Octavian Arsene
VIVirgil Ilian
Virgil Ilian
BCBogdan Cramariuc
Bogdan Cramariuc
Indoor positioning is one of the major topics in today's navigation and positioning research fields which is partially solved. The research community has not converged to a single, widely accepted solution that can achieve the desired accuracy at the required cost. Wireless Local Area Network (WLAN) based fingerprinting using Received Signal Strengths (RSS) is been considered as one solution for indoor positioning. This study structures this approach as supervised machine learning problem type where the target variable is the position and the features are the RSS values. There are compared the results obtained by from two analysis perspectives, regression and classification.
1 January 2019
Verified
Conference Paper
Closed access
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
OAO. Arsene
O. Arsene
MDM. Dragoi
M. Dragoi
ISI.D. Stanciu
I.D. Stanciu
IMI. Mocanu
I. Mocanu
In this paper, we describe an Ambient Assisted Living (AAL) scenario which includes a robotic platform employed in automatic data collection and assistance. As part of this scenario we propose new algorithms and solutions for indoor localization and manipulation. The former involves automatic data collection of WLAN RSS fingerprints by the robotic platform. The latter involves object manipulation with focus on grasping algorithms. Our new hybrid approach, where the localization problem is split into two stages, which are modelled as a classification and regression problem, proves to deliver the best positioning accuracy even when compared with deterministic approaches. At the same time, we also obtained improved results in the object grasping approach. The improvement is most pronounced for previously unseen data showing that our method is able to better generalize new environments.
1 October 2019
Verified
Conference Paper
Closed access
AAA.D. Andronescu
A.D. Andronescu
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
GTG.S. Tiplica
G.S. Tiplica
This paper proposes a solution for the most easily detectable type of cancer, skin cancer. Due to the ozone layer thinning, it was observed an increase in the number of skin lesion anomalies. Those anomalies are produced by the increasing UV radiation intensity. Good protection against sun is not always facile, mainly in the exposed regions of the skin like the face and the lower regions of the arms. Most types of cancers can be easily extracted via surgery if they are detected in one of the incipient development stages. Using a high resolution image, a classification algorithm can detect certain patterns in the nevi color, shape, veins or irregularities inside. Analyzing those aspects, a person can track their nevi development easily and intervene if it necessary. The originality of this paper consist in designing several convolutional neural networks, using the Python programming language, the Keras API and the Tensorflow framework alongside with proper dataset selection. The convolutional neural networks were trained using an open source dataset, with the purpose of formulating a diagnosis for new patients. The convolutional neural networks are provided with 7 classes of images, representing 7 types of skin lesions as input, and will output a diagnosis. The result of the research was a classifier with a convenient classification accuracy based on the dataset, and a medium accuracy based on new data.
1 June 2007
Verified
Conference Paper
Closed access
DNDumitru Iulian Nastac
Dumitru Iulian Nastac
PCPaul Cristea
Paul Cristea
The paper presents a flexible artificial neural network (ANN) model, in order to support modifications of a complex input-output function that describes the catalyst monitoring process of a multi-tube reactor. The goal is to obtain a good accuracy of the predicted data by using an optimal ANN architecture and well-suited delay vectors. The research targets the implementation of an adaptive system, which can be periodically retrained, in order to continuously learn the latest evolution of the catalyst process.