Browse Articles

Transfer Learning for Small and Different Datasets: Fine-Tuning A Pre-Trained Model Affects Performance

Gupta et al. | Oct 18, 2020

Transfer Learning for Small and Different Datasets: Fine-Tuning A Pre-Trained Model Affects Performance

In this study, the authors seek to improve a machine learning algorithm used for image classification: identifying male and female images. In addition to fine-tuning the classification model, they investigate how accuracy is affected by their changes (an important task when developing and updating algorithms). To determine accuracy, a set of images is used to train the model and then a separate set of images is used for validation. They found that the validation accuracy was close to the training accuracy. This study contributes to the expanding areas of machine learning and its applications to image identification.

Read More...

An explainable model for content moderation

Cao et al. | Aug 16, 2023

An explainable model for content moderation

The authors looked at the ability of machine learning algorithms to interpret language given their increasing use in moderating content on social media. Using an explainable model they were able to achieve 81% accuracy in detecting fake vs. real news based on language of posts alone.

Read More...

A machine learning approach for abstraction and reasoning problems without large amounts of data

Isik et al. | Jun 25, 2022

A machine learning approach for abstraction and reasoning problems without large amounts of data

While remarkable in its ability to mirror human cognition, machine learning and its associated algorithms often require extensive data to prove effective in completing tasks. However, data is not always plentiful, with unpredictable events occurring throughout our daily lives that require flexibility by artificial intelligence utilized in technology such as personal assistants and self-driving vehicles. Driven by the need for AI to complete tasks without extensive training, the researchers in this article use fluid intelligence assessments to develop an algorithm capable of generalization and abstraction. By forgoing prioritization on skill-based training, this article demonstrates the potential of focusing on a more generalized cognitive ability for artificial intelligence, proving more flexible and thus human-like in solving unique tasks than skill-focused algorithms.

Read More...

Utilizing meteorological data and machine learning to predict and reduce the spread of California wildfires

Bilwar et al. | Jan 15, 2024

Utilizing meteorological data and machine learning to predict and reduce the spread of California wildfires
Image credit: Pixabay

This study hypothesized that a machine learning model could accurately predict the severity of California wildfires and determine the most influential meteorological factors. It utilized a custom dataset with information from the World Weather Online API and a Kaggle dataset of wildfires in California from 2013-2020. The developed algorithms classified fires into seven categories with promising accuracy (around 55 percent). They found that higher temperatures, lower humidity, lower dew point, higher wind gusts, and higher wind speeds are the most significant contributors to the spread of a wildfire. This tool could vastly improve the efficiency and preparedness of firefighters as they deal with wildfires.

Read More...

Using data science along with machine learning to determine the ARIMA model’s ability to adjust to irregularities in the dataset

Choudhary et al. | Jul 26, 2021

Using data science along with machine learning to determine the ARIMA model’s ability to adjust to irregularities in the dataset

Auto-Regressive Integrated Moving Average (ARIMA) models are known for their influence and application on time series data. This statistical analysis model uses time series data to depict future trends or values: a key contributor to crime mapping algorithms. However, the models may not function to their true potential when analyzing data with many different patterns. In order to determine the potential of ARIMA models, our research will test the model on irregularities in the data. Our team hypothesizes that the ARIMA model will be able to adapt to the different irregularities in the data that do not correspond to a certain trend or pattern. Using crime theft data and an ARIMA model, we determined the results of the ARIMA model’s forecast and how the accuracy differed on different days with irregularities in crime.

Read More...

LawCrypt: Secret Sharing for Attorney-Client Data in a Multi-Provider Cloud Architecture

Zhang et al. | Jul 19, 2020

LawCrypt: Secret Sharing for Attorney-Client Data in a Multi-Provider Cloud Architecture

In this study, the authors develop an architecture to implement in a cloud-based database used by law firms to ensure confidentiality, availability, and integrity of attorney documents while maintaining greater efficiency than traditional encryption algorithms. They assessed whether the architecture satisfies necessary criteria and tested the overall file sizes the architecture could process. The authors found that their system was able to handle larger file sizes and fit engineering criteria. This study presents a valuable new tool that can be used to ensure law firms have adequate security as they shift to using cloud-based storage systems for their files.

Read More...

A meta-analysis on NIST post-quantum cryptographic primitive finalists

Benny et al. | Sep 21, 2024

A meta-analysis on NIST post-quantum cryptographic primitive finalists
Image credit: Benny et al. 2024

The advent of quantum computing will pose a substantial threat to the security of classical cryptographic methods, which could become vulnerable to quantum-based attacks. In response to this impending challenge, the field of post-quantum cryptography has emerged, aiming to develop algorithms that can withstand the computational power of quantum computers. This study addressed the pressing concern of classical cryptographic methods becoming vulnerable to quantum-based attacks due to the rise of quantum computing. The emergence of post-quantum cryptography has led to the development of new resistant algorithms. Our research focused on four quantum-resistant algorithms endorsed by America’s National Institute of Standards and Technology (NIST) in 2022: CRYSTALS-Kyber, CRYSTALS-Dilithium, FALCON, and SPHINCS+. This study evaluated the security, performance, and comparative attributes of the four algorithms, considering factors such as key size, encryption/decryption speed, and complexity. Comparative analyses against each other and existing quantum-resistant algorithms provided insights into the strengths and weaknesses of each program. This research explored potential applications and future directions in the realm of quantum-resistant cryptography. Our findings concluded that the NIST algorithms were substantially more effective and efficient compared to classical cryptographic algorithms. Ultimately, this work underscored the need to adapt cryptographic techniques in the face of advancing quantum computing capabilities, offering valuable insights for researchers and practitioners in the field. Implementing NIST-endorsed quantum-resistant algorithms substantially reduced the vulnerability of cryptographic systems to quantum-based attacks compared to classical cryptographic methods.

Read More...

Jet optimization using a hybrid multivariate regression model and statistical methods in dimuon collisions

Chunduri et al. | Jun 09, 2024

Jet optimization using a hybrid multivariate regression model and statistical methods in dimuon collisions
Image credit: Chunduri, Srinivas and McMahan, 2024.

Collisions of heavy ions, such as muons result in jets and noise. In high-energy particle physics, researchers use jets as crucial event-shaped observable objects to determine the properties of a collision. However, many ionic collisions result in large amounts of energy lost as noise, thus reducing the efficiency of collisions with heavy ions. The purpose of our study is to analyze the relationships between properties of muons in a dimuon collision to optimize conditions of dimuon collisions and minimize the noise lost. We used principles of Newtonian mechanics at the particle level, allowing us to further analyze different models. We used simple Python algorithms as well as linear regression models with tools such as sci-kit Learn, NumPy, and Pandas to help analyze our results. We hypothesized that since the invariant mass, the energy, and the resultant momentum vector are correlated with noise, if we constrain these inputs optimally, there will be scenarios in which the noise of the heavy-ion collision is minimized.

Read More...