Despite the emergence of artificial intelligence (AI) and machine learning (ML) in medicine and the resultant interest in predictive analytics in surgery, there remains a paucity of research on the actual impact of prediction models and their effect on surgeons’ risk assessment of post-surgical complications. This research evaluated how spinal surgeons predict post-surgical complications with and without additional information generated by a ML predictive model. The study was conducted in two stages. In the preliminary stage an ML prediction model for post-surgical complications in spine surgery was developed. In the second stage, a survey instrument was developed, using patient vignettes, to determine how providing ML model support affected surgeons’ predictions of post-surgical complications. Results show that support provided by a ML prediction model improved surgeons’ accuracy to correctly predict the presence or absence of a complication in patients undergoing spinal surgery from 49.1% to 54.8% (p=0.024). It is clear that predicting post-surgical complications in patients undergoing spinal surgery is difficult, for models and experienced surgeons, but it is not surprising that additional information provided by the ML model prediction was beneficial overall. This is the first study in the spine surgery literature that has evaluated the impact of a ML prediction model on surgeon prediction accuracy of post-surgical complications.

archives@tulane.edu / 1 / Sanjaya Lohani

This project is an experimentation and examination of contemporary computer vision and machine learning, with an emphasis on machine generated imagery and text, as well as object identification. In other words, this is a study of how computers and machines are learning to see and recognize the world. Computer vision is a kind of visual communication that we rarely think of as being designed. With an emphasis on written and visual research, this project aims to comprehend what exactly goes into the creation of machine generated imagery and artificial vision systems. I have spent the last couple of months looking through the lense of cameras, object identification apps and generative neural networks in order to try and understand how AI perceives reality. This resulted in a mixed media story about images and vision, told through the perspective of a fictional AI character. Visit ​www.maybe-probably.com​ to view the project.

Music has tremendous cultural and commercial significance for people the world over. It is one of the oldest human inventions and is among the most popular consumption activities on the planet. The music industry is also of great economic importance with 19 billion dollars in revenue worldwide in 2019. Despite music’s importance and significance, little work has been devoted to understanding what makes some types of music more popular than others or on the implications of success on artists’ subsequent productivity. Earlier studies have investigated psychological and economic aspects of music, but marketing as a field has devoted little attention to understanding the drivers of musical success and the dynamics of the music industry. In this dissertation, I leverage modern Bayesian non-parametric approaches, machine learning, and novel data to study the dynamic drivers of musical success and the implications of that success. The dissertation is composed of two essays devoted to investigating these complementary questions. In the first essay, I examine the dynamics of success of albums over the last fifty years. I then leverage the results to construct well-balanced playlists that will appeal to different generations of music listeners. My empirical investigation is based on a novel dataset I collected from diverse online sources. The dataset is comprised of albums' movements up and down Billboard magazine’s annual Top 200 lists of albums, marketing and standard descriptors of the albums such as genre and artist popularity, acoustic descriptors of the albums' tracks such as the songs’ acoustic fingerprints, and user-generated tags describing the albums’ and songs’ consumption context and the experience perceived by listeners. I develop a novel Bayesian non-parametric model that fuses the diverse data modalities and predicts the dynamic patterns of musical success over the years. The model generates results regarding how musical acoustic qualities and genres have waxed and waned in popularity over time. It also uses tags listeners generate online to uncover themes that categorize albums in terms of sub-genres, consumption contexts, emotions, evocation of nostalgia, and other aspects of the musical experience. The model yields insightful results about the evolution of album success in the music industry. These insights are relevant to artists and music professionals who recommend albums, design new releases, and construct well-balanced playlists aimed at various generations of listeners. The second essay is devoted to quantifying the effects of winning the Grammy for Best New Artist on artists’ productivity and musical variety. The causal identification strategy is based on comparing subsequent outcomes in terms of both productivity and diversification of musical styles and elements winners of and contenders for the award. This strategy allows the model to control for ability bias and improves confidence in the estimated causal effects. The study is based on a dataset I collected from diverse online sources that spans the entire history of the Best New Artist award and contains integral album discographies of the nominees, most of their released songs, and their acoustic descriptors. I use a two-way fixed effects approach to measure the causal effect of the award and incorporate heterogeneity in the treatment effects. The results yield interesting insights into positive effects of the award on productivity. Interestingly, my investigation also reveals that the effects of winning the award are heterogeneous in terms of gender and that male solo singers benefit more than female solo singers and groups, male groups, and mixed-gender groups. In contrast, winning the award does not affect artistic variety on average, though winners tend to explore new artistic dimensions that are congruent with their musical specialties than contenders do.

The objective of this work is to explore the intrusion detection prob- lem and create simple rules for detecting specific intrusions. The intrusions are explored in the realistic CSE-CIC-IDS2018 dataset. First, the dataset is analyzed by computing appropriate statistics and visualizing the data. In the data visu- alization various dimensionality reduction methods are tested. After analyzing the dataset the data are normalized and prepared for the training. The training process focuses on feature selection and finding the best model for the intrusion detection problem. The feature selection is also used for creating rules. The rules are extracted from an ensemble of Decision Trees. At the end of this work, the rules are compared to the best model. The experiments demonstrate that the simple rules are able to achieve similar results as the best model and can be used in a rule-based intrusion detection system or be deployed as a simple model. 1

NNBAR is a proposed experiment for the European Spallation Source. Thegoal of the experiment is to observe the transformation n −  ̄n. Currently a cutbased analysis is used to select signal events and discriminate against cosmic raybackground. To further increase the signal efficiency machine learning was used.Most machine learning algorithms resulted in a higher signal efficiency at the costof lowering the background rejection. However using the Linear DiscriminantAnalysis resulted in a new signal efficiency of 94% whilst having a predictedbackground rejection of roughly 100%. These results show that machine learningis a promising tool for increasing the signal efficiency at NNBAR.

Indiana University-Purdue University Indianapolis (IUPUI) / Mutual learning is an emerging field in intelligent systems which takes inspiration from naturally intelligent agents and attempts to explore how agents can communicate and coop- erate to share information and learn more quickly. While agents in many biological systems have little trouble learning from one another, it is not immediately obvious how artificial agents would achieve similar learning. In this thesis, I explore how agents learn to interact with complex systems. I further explore how these complex learning agents may be able to transfer knowledge to one another to improve their learning performance when they are learning together and have the power of communication. While significant research has been done to explore the problem of knowledge transfer, the existing literature is concerned ei- ther with supervised learning tasks or relatively simple discrete reinforcement learning. The work presented here is, to my knowledge, the first which admits continuous state spaces and deep reinforcement learning techniques. The first contribution of this thesis, presented in Chapter 2, is a modified version of deep Q-learning which demonstrates improved learning performance due to the addition of a mutual learning term which penalizes disagreement between mutually learning agents. The second contribution, in Chapter 3, is a presentation work which describes effective communication of agents which use fundamentally different knowledge representations and systems of learning (model-free deep Q learning and model- based adaptive dynamic programming), and I discuss how the agents can mathematically negotiate their trust in one another to achieve superior learning performance. I conclude with a discussion of the promise shown by this area of research and a discussion of problems which I believe are exciting directions for future research.

This thesis work, conducted at Telenor Sweden, aims to build a model that would try to accurately predict the resolution time of Priority 4 Trouble Tickets. (Priority 4 trouble tickets are those tickets that get generated more often-e in higher volumes per month). It explores and investigates the possibility of applying Machine Learning and Deep Learning techniques to trouble ticket data to find an optimal solution that performs better than the current method in place (which is explained in Section 3.5). The model would be used by Telenor to inform the end-users of when the networks team expects to resolve the issues that are affecting them.

The computational nature of soft computing e.g. machine learning and AI systems have been hidden by seamless interfaces for almost two decades now. It has led to the loss of control, inability to explore, and adapt to needs and privacy at an individual level to social-technical problems on a global scale. I propose a soft machine - a set of cohesive design patterns or ‘seams’ to interact with everyday ‘black-box’ algorithms. Through participatory design and tangible sketching, I illustrate several interaction techniques to show how people can naturally control, explore, and adapt in-context algorithmic systems. Unlike existing design approaches, I treat machine learning as playful ‘design material’ finding moments of interplay between human common sense and statical intelligence. Further, I conceive machine learning not as a ‘technology’ but rather as an iterative training ‘process’, which eventually changes the role of user from a passive consumer of technology to an active trainer of algorithms.

Compositional data refers to simplex-valued data, or equivalently, nonnegative vectors whose totals are uninformative. This data modality is of relevance across several scientific domains. A classical example of compositional data is the chemical composition of geological samples, e.g., major-oxide concentrations. A more modern example arises from the microbial populations recorded using high-throughput genetic sequencing technologies, e.g., the gut microbiome. This dissertation presents a set of methodological and theoretical contributions that advance the state of the art in the analysis of compositional data. Our work can be divided along two categories: problems in which compositional data represents the input to a predictive model, and problems in which it represents the output of the model. For the first class of problems, we build on the popular log-ratio framework to develop an efficient learning algorithm for high-dimensional compositional data. Our algorithm runs orders of magnitude faster than competing alternatives, without sacrificing model quality. For the second class of problems, we define a novel exponential family of probability distributions supported on the simplex. This distribution enjoys attractive mathematical properties and provides a performant probability model for simplex-valued outcomes. Taken together, our results constitute a broad contribution to the toolkit of researchers and practitioners studying compositional data.