Source author record

Iury Bessa

Iury Bessa appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Systems and Control eess.SY math.OC Software Engineering Artificial Intelligence Machine Learning math.DS

Catalog footprint

What is connected

4works

7topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2022arXiv

CEG4N: Counter-Example Guided Neural Network Quantization Refinement

Neural networks are essential components of learning-based software systems. However, their high compute, memory, and power requirements make using them in low resources domains challenging. For this reason, neural networks are often quantized before deployment. Existing quantization techniques tend to degrade the network accuracy. We propose Counter-Example Guided Neural Network Quantization Refinement (CEG4N). This technique combines search-based quantization and equivalence verification: the former minimizes the computational requirements, while the latter guarantees that the network's output does not change after quantization. We evaluate CEG4N~on a diverse set of benchmarks, including large and small networks. Our technique successfully quantizes the networks in our evaluation while producing models with up to 72% better accuracy than state-of-the-art techniques.

preprint2021arXiv

Evolving Fuzzy System Applied to Battery Charge Capacity Prediction for Fault Prognostics

This paper addresses the use of data-driven evolving techniques applied to fault prognostics. In such problems, accurate predictions of multiple steps ahead are essential for the Remaining Useful Life (RUL) estimation of a given asset. The fault prognostics' solutions must be able to model the typical nonlinear behavior of the degradation processes of these assets, and be adaptable to each unit's particularities. In this context, the Evolving Fuzzy Systems (EFSs) are models capable of representing such behaviors, in addition of being able to deal with non-stationary behavior, also present in these problems. Moreover, a methodology to recursively track the model's estimation error is presented as a way to quantify uncertainties that are propagated in the long-term predictions. The well-established NASA's Li-ion batteries data set is used to evaluate the models. The experiments indicate that generic EFSs can take advantage of both historical and stream data to estimate the RUL and its uncertainty.

preprint2021arXiv

Incremental Learning and State-Space Evolving Fuzzy Control of Nonlinear Time-Varying Systems with Unknown Model

We present a method for incremental modeling and time-varying control of unknown nonlinear systems. The method combines elements of evolving intelligence, granular machine learning, and multi-variable control. We propose a State-Space Fuzzy-set-Based evolving Modeling (SS-FBeM) approach. The resulting fuzzy model is structurally and parametrically developed from a data stream with focus on memory and data coverage. The fuzzy controller also evolves, based on the data instances and fuzzy model parameters. Its local gains are redesigned in real-time -- whenever the corresponding local fuzzy models change -- from the solution of a linear matrix inequality problem derived from a fuzzy Lyapunov function and bounded input conditions. We have shown one-step prediction and asymptotic stabilization of the Henon chaos.

preprint2014arXiv

SMT-Based Bounded Model Checking of Fixed-Point Digital Controllers

Digital controllers have several advantages with respect to their flexibility and design's simplicity. However, they are subject to problems that are not faced by analog controllers. In particular, these problems are related to the finite word-length implementation that might lead to overflows, limit cycles, and time constraints in fixed-point processors. This paper proposes a new method to detect design's errors in digital controllers using a state-of-the art bounded model checker based on satisfiability modulo theories. The experiments with digital controllers for a ball and beam plant demonstrate that the proposed method can be very effective in finding errors in digital controllers than other existing approaches based on traditional simulations tools.