GPT-2 was introduced in 2019 but we can learn how the fundamentals of transformer training and inference work by rebuilding GPT-2 in PyTorch.
GPT-2 was introduced in 2019 and primarily built within the TensorFlow framework. Let's take a look at how we can reproduce GPT-2 with PyTorch while following the methodology outlined in the accompanying GPT-2 research paper.
Prompt engineering language models lacks necessary refinement especially for enterprise solutions. Optimize your language models with DSPy!
Optimizing language models is an important task as organizations integrate generative AI capabilities into their business processes. Relying on manual prompt engineering introduces risk, biases, and other problems. Continue reading to learn how to mitigate these risks with DSPy.
LLaVA, among other multimodal AI models, leaps from standalone large language models and allows us to engage with AI in novel ways.
Understanding our world often requires interpreting stimuli from multiple sources including through vision or text. LLaVA accomplishes this by combining a vision encoder with a large language model. Continue reading to learn how to leverage multimodal AI with Databricks and LLaVA! 🌋
Using Ultralytics YOLOv8 and Databricks, we can build and deploy an AI model to reliably detect objects.
For many AI-powered use cases, object detection emerges as a critical capability. Examples include driving cars, performing surgery, or identifying product defects. Continue reading to learn how to leverage Databricks and YOLOv8 to detect objects! 🤖
OpenAI emerged as a leader in generative AI and is helping enterprises modernize their processes at a rapid pace. Get started with generative AI, OpenAI, and more!
Generative AI is a compelling technology whether you are an artist looking for inspiration or a CEO looking to innovate. OpenAI captured the world's imagination after introducing ChatGPT - how can you get started with OpenAI's platform? Continue reading to learn more.
Edge AI emerges at the intersection of AI, microcontrollers, and process automation. Jump start your understanding of these topics and more!
Deploying AI in the field challenges organizations to this day. Thanks to innovation, constraints such as limited tools, employee skills, and hardware specifications are becoming blockers of the past. Continue reading to learn more.
A data scientist's experience at The Firm.
Data science, ML, and AI are changing organizations. What does this mean for consultants and clients? Continue reading to learn about my experience as a data scientist at McKinsey.
Kedro, an open-source Python framework, helps data science teams create maintainable data pipelines.
Folks have compared data to oil in the past, and while the metaphor may be tired, the process of building pipelines in both cases consumes valuable time and resources, and often results in a mess. Read more to learn about Kedro!
Databricks is the Data + AI company and the provider of a leading data science and machine learning platform.
Ingesting, cleaning, and analyzing data on Databricks accelerates your data science and machine learning projects by leveraging the underlying data lake and Spark engine. Continue reading to get started with Databricks.
Hyperopt is a popular tool for optimizing hyperparameters in machine learning and data science.
Hyperopt enables you to perform optimization in parallel over a defined search space to find the optimal parameters. Continue reading to get started with Hyperopt.
Optimization with MATLAB and GAMS can aid decision-makers in the for-profit, non-profit, and public sectors.
Using optimization methods requires structuring problems, identifying objectives, and imposing constraints given a problem context. Continue reading to learn how to solve optmization problems in MATLAB and GAMS.
Probability distributions are used throughout quantitative fields where uncertainty or randomness are involved.
Probabilistic thinking requires knowledge of probability distributions and the relevant statistics associated with each. Understanding these distributions can help you identify opportunities to leverage one. Continue reading to learn about the core concepts and essential distributions with examples in Python.
SHAP or Shapley Additive Explanations helps you understand ML models.
The SHAP framework unifies the methods used to interpret and explain machine learning models. This post helps interpret and explain SHAP. Read this post to start getting into SHAP (with both high-level explanation and python example).
Time series analysis is a powerful technique to generate forecasts with seasonal data.
Data sets contain noise but with high-powered or elegant data science, the relevant signal can be extracted. One key technique for analysis of real-world data (primarily focused on forecasting) is time-series analysis. A popular time-series forecasting procedure is Facebook's open-source Prophet procedure. Prophet is implemented in both R and Python.
An expression of gratitude and summary of experience from one of Carnegie Mellon's latest graduates.
Carnegie Mellon University's Heinz College offered a unique opportunity to jointly study public policy, management, and data analytics. The coursework covers topics ranging from machine learning, deep learning, econometrics, and optimization to organizational design and decision-making. Continue reading to learn why I am glad I enrolled.
Using only ingredients, learn how to classify recipes into cuisines using Python, classification methods, clustering methods, and Sci-kit Learn.
While seemingly a trivial task, classifying recipes into cuisines and understanding how to interpret clustering and classification results can help you creatively answer other questions. Continue reading to learn how.
Suspicious connection requests are a common occurrence for Linux system administrators. From who and where are these suspicious requests originating?
Cyber attacks are becoming more common as more data are stored in digital form. Suspicious, and likely malicious, users are requesting access to unauthorized resources while hoping for vulnerable networks or systems. Continue reading to learn how to apply descriptive analytics to uncover who is generating these suspicious requests.
From concept to code: introduce yourself to neural networks and deep learning.
Neural networks are a growing area of research and are being applied to new problems every day. The classic examples are image classification, facial recognition, and self-driving cars. The future is uncertain, but there is a high probability that neural networks in some form will play a critical role in shaping it. Continue reading to learn about the core concepts and walk through an example in Python.
Python and R are popular in data science. Which is best?
Data science is a broad and deepening field but one divisive question still remains. In this post, I compare and contrast Python and R for use in different data science tasks. Continue reading to learn which one emerges triumphant.
Learn how to build a shiny dashboard in R to help users analyze, visualize, and understand their data.
Shiny dashboards provide a simple and fast way to analyze and visualize data. Whether performing exploratory data analysis or building a robust tool for your client's executives, Shiny dashboards aid the data science process. Continue reading to walk through an example of constructing an R Shiny Dashboard.
Data science for good: my internship with the ACLU in NYC
Data science can change how any organization operates not just Facebook or Amazon, but even the American Civil Liberties Union - an organization 10 times older than Facebook. Continue reading to learn about my experience interning with the ACLU's data science team in New York.
Can data science indicate what factors affect the rate of criminal or violent recidivism? (Hint: Yes)
Around the United States, municipalities have turned to risk assessment instruments (RAIs) to help judges determine which individuals to release on bail and which ones to keep in custody. The risk assessment process varies based on the specific instrument used but many rely on criminal recidivism data sets. These data sets typically contain various demographic indicators (age, race, gender, etc.) and also criminal history (charges, juvenile record, etc.).
Broward County, Florida, has turned to the use of one of the most popular RAIs today: COMPAS or the Correctional Offender Management Profiling for Alternative Sanctions tool. COMPAS assesses individuals based on criminal history and social profiling to categorize an individual as low, medium, or high risk. This tool, however, was not developed using the Broward County data set which may lead to poor performing predictions for individuals from Broward County, Florida. In this post we construct an RAI, compare to COMPAS and discuss findings.
How do factors (besides gender) affect the gender income gap?
The issue of gender-income disparity is not new - it is nuanced. In order to answer these questions, we will rely on the National Longitudinal Survey of Youth, 1979 cohort, data set (abbreviated ‘NLSY79’). To draw strong conclusions, we must evaluate the data set provided - is it accurate, relevant, and useful for drawing statistical conclusions? Once we summarize the data, we can discuss methodology - how should we approach the data, what variables should we consider, what techniques are appropriate? Third, we openly discuss findings about the sampled individuals and attempt to infer relationships about the income difference (if any) between men and women in the larger population based on other factors. Lastly, we end with a discussion of the relevancy and signicance of our findings given the context of the survey data available and the methodology used.
A simple 2x2 table can contain insightful metrics that enhance your decision-making.
Understanding the confusion matrix is an important step in statistics, machine learning, or any other field where predictions or classifications are common. The confusion matrix is a type of contingency table with two dimensions that reveal how well a predictive model performs when the outcomes are known. Additionally, when associated costs of incorrect positive and negative guesses differ, the trade-offs can be optimized. Do you know the difference between Sensitivity, Specificity, Recall, Precision, True Positive Rate, and Positive Predictive Value?
How Hugo and I created this new personal blog.
Read the story about how I replaced my old website (from 2009) with a new one that is more modern, features easy publishing, and requires low maintenance. Along the way, I include tips and tricks that will help you create your own website.
© Ben Hayes, Theme by: HTML5 UP.
