Mariana Meraz

Chapter 7 compares TensorFlow and PyTorch, two well-known deep learning frameworks. Google-developed TensorFlow boasts a bigger ecosystem and is well known for its performance and deployment via tools like TensorFlow Serving and TensorFlow Lite. Facebook-developed PyTorch is increasingly popular in research because of its Pythonic, intuitive API and dynamic computation graphs, which make coding and debugging easier. While TensorFlow has improved ease of use with version 2.0, PyTorch remains the go-to choice for experimentation and learning. Although they are machine learning tools, they offer insight even for non-technical students in fields like accounting. As data and automation continue to revolutionize the accounting industry, having a general understanding of how machine learning tools work can be helpful. For example, they can help with tasks like fraud detection, financial forecasting, or cost analysis using predictive models. After graduation, such knowledge can fill the gap be...

In this chapter, we were introduced to PyTorch, a dynamic and beginner-friendly machine learning library created by Facebook's AI Research Lab. The instructional style focused on hands-on learning, the same style championed by fast.ai, where students were encouraged to get their hands dirty with creating actual tools instead of being bogged down by theory. We began by learning about tensors, multi-dimensional arrays that can be computed on GPUs for speed. We went over basic tensor operations like addition and multiplication, and immediately applied them to a real task: housing price prediction on the Boston Housing dataset. We gradually prepared the data, defined a simple neural network (SimpleNet), trained it using Mean Squared Error (MSE) loss and Stochastic Gradient Descent (SGD) optimizer, and evaluated its performance on unseen test data. The chapter also included a concise Q&A section that clarified such key concepts as epochs, the need for separate training and test sets...

Data are at the forefront of business decision-making today, enabling organizations to work smarter and stay ahead. Firms use information in many different ways—such as marketing, forecasting, segmentation, supply chain management, and risk management. With data-driven marketing, companies can create tailor-made campaigns based on customer behavior and preferences, increasing conversion rates and customer loyalty. Predictive analytics facilitates companies to predict future trends and needs, improving planning and saving on costs. Customer segmentation contributes to personalization by grouping customers into demographics and behavior clusters, allowing for targeted retention and engagement. Operations are enhanced through data by streamlining supply chains by reducing complexity in logistics, inventory management, and making deliveries more efficient. As an example, real-time tracking and performance metrics allow companies to act in real-time if there is delay or interruption. Anothe...

Chapter 5 provides a good foundation in probability and statistics, both of which are not only essential in data science but also very relevant to accounting and financial analysis. Mean, median, variance, and standard deviation are methods of descriptive statistics that allow accountants to condense financial information quickly and briefly. For example, in considering costs or revenues over time, these figures may be utilized in order to observe patterns; typical monthly costs or the level of fluctuation of sales. Such information can facilitate budgeting decisions and variance analysis, which underlie accounting tasks. The presentation to probability distributions, including the normal distribution, has real-world applications within accounting risk analysis and forecasting. Actually practiced, accountants use these distributions as tools for fitting uncertainties in forecasting future cash flows, computing bad debt, or sensitivity analysis. Knowledge about how the data shifts in te...

Chapter 4 is devoted to deep learning and neural networks, which explain how computers learn from data similar to humans. Deep learning, a subset of machine learning, allows systems to learn patterns and make decisions without programming. It is especially helpful in dealing with huge amounts of complex data such as images, sounds, and text. Deep learning is spurred by neural networks, which mimic the human brain by consisting of layers of interconnected "neurons." Neural networks are capable of learning relationships between data, and different types including CNNs for pictures or RNNs for sequences are utilized depending on the task. As a student of accounting, I can see how deep learning could improve the majority of the field. For instance, neural networks can be trained to alert fraud by going through financial transactions and unusual patterns. That would be immensely useful to internal audit or financial oversight. Deep learning-based systems also digitize paper docume...

Linear algebra may seem far removed from accounting at first glance, but it possesses tools of utility that can be used to assist us in analyzing and interpreting financial data more effectively. At its most fundamental level, accounting is all about manipulating, managing, and interpreting numerical data—exactly the kind of activity linear algebra is well-suited to. Vectors and matrices are useful in financial modeling, budgeting, forecasting, and even evaluating multiple variables within departments or business segments. For example, you can graph a number of financial measures ( revenue, spending, and net profit) of various branches or time frames in a matrix and calculate across the matrices to discover patterns, averages, or trends. In practice, linear algebra can minimize and streamline much accounting work when paired with programming such as Python and NumPy. Calculations like year-over-year comparisons, cost allocations, or even data normalization between accounts can be autom...