Loading...
Publication
Avaliação da Representação Neuronal Implícita para Armazenamento de Dados em ADN
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.07 MB | Adobe PDF |
Authors
Abstract(s)
A crescente demanda por soluções de armazenamento de alta densidade e longa duração tem colocado o ADN como uma alternativa promissora, graças à sua capacidade de armazenar grandes volumes de dados de forma compacta e durável. Entre os diversos tipos de dados, as imagens destacam-se pela sua importância em áreas como medicina, arte, comunicação, entre outras, tornando essencial o desenvolvimento de métodos eficientes para sua representação em ADN. A natureza complexa das imagens exige abordagens especializadas que considerem suas particularidades, como redundância espacial e correlação entre pixeis, além das restrições bioquímicas do ADN. Inicialmente, as abordagens de codificação em ADN tratavam todos os tipos de dados de forma homogénea, sem adaptações específicas para imagens. Com o tempo, codificadores clássicos de imagem, como JPEG, foram adaptados para atender às restrições do ADN, e métodos baseados em inteligência artificial, como o COOL-CHIC, ganharam destaque por sua eficiência na compressão e preservação da qualidade visual. Inspirado nesses avanços, o HiDNA surgiu como um codificador inovador, combinando técnicas de IA com as restrições bioquímicas do DNA para representar imagens de forma robusta e eficiente. Neste trabalho, exploramos o uso de representações neuronais implícitas na codificação de imagens com vista ao seu armazenamento em Ácido Desoxirribonucleico (ADN), avaliando sua capacidade de compressão, robustez fidelidade na recuperação de dados. Métodos como o COOL-CHIC demonstram ganhos significativos na relação taxa-distorção e na redução do custo computacional, apontando para o potencial de tornar o armazenamento genético escalável e sustentável. A nossa contribuição visa avançar nesta direção, adaptando codificadores, tal como efetuado pelo HiDNA, ainda mais adaptados às necessidades especificas das imagens, abrindo caminho para aplicações práticas e inovadoras no campo do armazenamento em ADN.
The growing demand for high-density, long-term storage solutions has made DNA a promising alternative, thanks to its ability to store large volumes of data in a compact and durable way. Among the various types of data, images stand out for their importance in areas such as medicine, art, communication, among others, making it essential to develop efficient methods for their representation in DNA. The complex nature of images requires specialized approaches that consider their particularities, such as spatial redundancy and correlation between pixels, in addition to the biochemical constraints of DNA. The complex nature of images requires specialized approaches that consider their particularities, such as spatial redundancy and correlation between pixels, in addition to the biochemical constraints of DNA. Initially, DNA encoding approaches treated all types of data homogeneously, without specific adaptations for images. Over time, classic image encoders, such as JPEG, were adapted to meet DNA constraints, and methods based on artificial intelligence, such as COOL-CHIC, gained prominence for their efficiency in compression and preservation of visual quality. Inspired by these advances, HiDNA emerged as an innovative encoder, combining AI techniques with the biochemical constraints of DNA to represent images robustly and efficiently. In this work, we explore the use of implicit neural representations in image encoding for storage in deoxyribonucleic acid (DNA), evaluating their compression capacity, robustness, and fidelity in data retrieval. Methods such as HiDNA and COOL-CHIC have demonstrated significant gains in rate-distortion and reduced computational cost, pointing to the potential to make genetic storage scalable and sustainable. Our contribution aims to advance in this direction by developing encoders more adapted to the specific needs of images, paving the way for practical and innovative applications in the field of DNA storage.
The growing demand for high-density, long-term storage solutions has made DNA a promising alternative, thanks to its ability to store large volumes of data in a compact and durable way. Among the various types of data, images stand out for their importance in areas such as medicine, art, communication, among others, making it essential to develop efficient methods for their representation in DNA. The complex nature of images requires specialized approaches that consider their particularities, such as spatial redundancy and correlation between pixels, in addition to the biochemical constraints of DNA. The complex nature of images requires specialized approaches that consider their particularities, such as spatial redundancy and correlation between pixels, in addition to the biochemical constraints of DNA. Initially, DNA encoding approaches treated all types of data homogeneously, without specific adaptations for images. Over time, classic image encoders, such as JPEG, were adapted to meet DNA constraints, and methods based on artificial intelligence, such as COOL-CHIC, gained prominence for their efficiency in compression and preservation of visual quality. Inspired by these advances, HiDNA emerged as an innovative encoder, combining AI techniques with the biochemical constraints of DNA to represent images robustly and efficiently. In this work, we explore the use of implicit neural representations in image encoding for storage in deoxyribonucleic acid (DNA), evaluating their compression capacity, robustness, and fidelity in data retrieval. Methods such as HiDNA and COOL-CHIC have demonstrated significant gains in rate-distortion and reduced computational cost, pointing to the potential to make genetic storage scalable and sustainable. Our contribution aims to advance in this direction by developing encoders more adapted to the specific needs of images, paving the way for practical and innovative applications in the field of DNA storage.
Description
Keywords
Armazenamento em Adn Co-
Dec Hidna Compressão de Dados Cool-Chic Representação Neuronal Implícita Taxa-Distorção.