U Bremen Research Alliance’s Post

Exploring the Frontiers of Data Science: Dive into Sechel Ventures Partners LLC enlightening discussion with Dr. paul laskowski, Adjunct Assistant Professor at the University of California, Berkeley. Our latest article offers an in-depth look at the evolving field of data science, with Dr. Laskowski sharing his invaluable insights on emerging trends and the future challenges that professionals in the field can expect to face. Stay Ahead in Data Science: Whether you're a seasoned expert or a budding data scientist, this feature provides practical advice and foresight to help you navigate the complexities of this dynamic industry. Explore essential tips and strategies from a leading academic in the field. #DataScience #Analytics #TechTrends #UCBerkeley

A harsh truth about data science… The core of a data scientist's role is deciphering data to generate value and inform decisions. It’s not necessarily deploying complex ML/AI techniques—though these can be tools in your arsenal. A common trap for emerging data scientists is the rush to apply advanced methods, when tools drive objectives rather than the reverse. Often, simple analyses suffice. Time series Contingency tables Histograms For instance, while building a complex model like a multi-layer CNN might showcase technical prowess, its business application—like identifying potential up-sell customers—could often be achieved with simpler models, more quickly and efficiently. The skill of selecting the right tool for the problem and converting data insights into business actions is rarely emphasized in academic settings, leaving a gap in practical application knowledge. The focus should shift towards fostering these critical decision-making and strategic translation abilities right from the start.

My next class approaches! This time, I'm teaching critical perspectives on data science. If you've wanted to tune your data BS-meter or if you're a practitioner who's been wanting (or told) to read some humanities, this is the class for you. I'm aiming to give people general critical tools. There is so much amazing and important writing on ethics and society in this space, but we're going to focus more historically and philosophically so that you can think for yourself when something new lands. Some topics we'll address: How exactly do data scientists convert data into business value? What's the relationship between data science and statistics? What is machine learning, and what does it have to do with artificial intelligence? What good are predictions if these "black boxes" don't also give us theories? What are the fundamental limitations of any data-sensitive algorithmic system? And of course, what even is data science? https://lnkd.in/ekvqHNbE Meets virtually, starts in a month. Signups are already almost full, but if the waitlist grows enough, we may open another section meeting in-person in NYC. (If you know of any spaces that might be willing to host us, let me know!) Feel free to spread the word or check out our other classes! https://lnkd.in/dp9xmt7a PS, my last class went really well. As an educator, "I learned so much!" is the best compliment.

INFORMS Journal on Data Science Welcomes Submissions INFORMS Journal on Data Science, with its first issue published in 2021, is a new addition to the INFORMS publication family. IJDS serves both engineering data science and business data science communities. I am the new Editor-in-Chief for IJDS and I am writing to welcome your submissions to this relatively new journal. 1. All new submissions to IJDS now follow a double-blind review process. When you make a submission, please remove all author-identifying information from both the main manuscript and supplemental materials (if applicable), including those in the acknowledgement section. 2. For new submissions on and after 1/1/2025, authors are no longer required to use Code Ocean to submit data and code. We move the data/code sharing to the time when a paper is accepted, not at the time of submission. At submission, authors need to submit a Data/Code Disclosure Form for acknowledging the reproducibility requirement, including data/code sharing, when the paper is accepted for publication.  3. I acknowledge that IJDS's review time was long, but we are making effort to shorten the review time. Our goal is to have 90% of the submissions reviewed and decisioned within 90 days and ensure that no paper review goes beyond 180 days. When these goals are met, one could expect the average review time to be around 45 days. In the past few weeks, I met one-on-one with each editor on the IJDS editorial board. All editors are on the same page, offering their best to meet these review time goals and provide quality reviews. It would be great if potential authors could go over the revised submission guidelines at https://lnkd.in/evD-eDc4 before submission. One can also find the current editorial board at https://lnkd.in/eN9nfg-5. If you think there is anything we can do to make paper submissions less burdensome, please feel free to reach out to me or any of our editors.

Thanks for featuring us in your publication, Data & AI Magazine!

As a Visiting Student at the Bocconi Institute for Data Science and Analytics, I recently presented the paper "Hierarchical Mixture Modeling with Normalized Inverse-Gaussian Priors" by Lijoi, Mena, and Prünster (2005) to peers and faculty. The introduction of the Dirichlet Process (DP) by Ferguson (1973) provided the missing link for the application of the Bayes-Laplace paradigm to nonparametric settings. Since then, research in the field of Bayesian nonparametrics has grown significantly. Blackwell (1973) showed a limitation of the DP: its almost sure discreteness, which represents a problem when data come from a continuous distribution. Lo (1984) addressed this issue by introducing the mixture of Dirichlet process, a model for density estimation. Lijoi, Mena, and Prünster proposed the Normalized Inverse-Gaussian (N-IG) prior as an alternative prior in mixture modeling. This prior, while being almost surely discrete like the DP, offers some key advantages. My presentation included an outline of the N-IG distribution and its key properties, including the additivity property inherited from the Inverse-Gaussian distribution. The N-IG lacks the conjugacy property, but thanks to modern computational techniques, this does not represent a limitation anymore. I also provided a description of the N-IG process, defined via its family of finite-dimensional distributions. I then focused on one of the most interesting features of the N-IG process: its predictive distribution. Unlike the DP, which assigns mass to each observation based solely on the frequency of observed values, the N-IG process's predictive distribution accounts for the entire clustering structure. This mechanism results in a clustering behavior that adapts more effectively to the complexity of the observed data. The authors also provided an explicit expression for the distribution of the number of distinct components in a sample, and this quantity is particularly relevant as it provides a prior for the number of components in the mixture. The presentation concluded with a review of key results from the authors' analysis, highlighting the N-IG process's performance and advantages in hierarchical mixture models.

🚀 Excited to dive deep into data science with Berkeley University's Data 8 course! Currently immersed in the "Computational and Inferential Thinking: The Foundations of Data Science" textbook. 📘✨ This week's focus has been on understanding the essence of data science and its importance. Key takeaways include: What is Data Science? It's about drawing useful conclusions from large and diverse data sets through exploration, prediction, and inference. Exploration identifies patterns, prediction makes informed guesses, and inference quantifies certainty. Tools include visualizations, descriptive statistics, machine learning, optimization, and statistical tests. Why Data Science? It reduces uncertainty in decision-making across various industries by leveraging large data sets and computational tools. Data-driven decision making has transformed industries like finance, advertising, and real estate. The field promotes critical thinking supported by data, enabling precise, reliable, and quantitative arguments. Data science combines statistics and computing to analyze real-world data sets, including text, images, videos, and sensor readings. It empowers individuals to apply these techniques in their work, scientific endeavors, and personal decisions. The goal is to make data reasoning accessible to everyone, fostering a society capable of tackling unanswered questions and challenges with data-driven insights. Looking forward to applying these skills to projects and sharing more insights as I progress through the course. If you're also into data science or taking the same course, let's connect and learn together! #DataScience #Data8 #BerkeleyUniversity #ComputationalThinking #StatisticalInference #PythonProgramming #MachineLearning #DataAnalysis #ContinuousLearning

Sometime before and during the COVID-19 pandemic, we started meeting every week to discuss: what does it mean to be responsible in data science projects? We thought of responsibility as a series of dimensions and scopes and proposed a tool which we named: TAPS Responsibility matrix. After years of work and refinements, the work is now available at the Journal of Responsible Innovation. Thanks to all the work and efforts of all the co-authors Remzi Celebi Chang Sun Parveen Kumar 🟥 Linda Rieswijk Michael Erard Kody Moodley Dr. Arif YILMAZ Michel Dumontier as well as the many experts who gave their time and expert advise. https://lnkd.in/e_8RUrNE #responibleai #datascience

✨ 𝗜𝗻𝘁𝗲𝗿𝗱𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗮𝗿𝘆 𝘀𝗸𝗶𝗹𝗹𝘀 𝗮𝗿𝗲 𝗸𝗲𝘆 𝗶𝗻 𝗱𝗮𝘁𝗮 𝘀𝗰𝗶𝗲𝗻𝗰𝗲. The most innovative solutions often emerge when insights from diverse fields come together: 🔍 𝗣𝘀𝘆𝗰𝗵𝗼𝗹𝗼𝗴𝘆: Helps us design user-centric models and improve sentiment analysis by understanding human behavior. 📊 𝗘𝗰𝗼𝗻𝗼𝗺𝗶𝗰𝘀: Offers frameworks for decision-making and optimization, enabling smarter business strategies. 👥 𝗦𝗼𝗰𝗶𝗼𝗹𝗼𝗴𝘆: Provides insights into group dynamics, aiding in the analysis of social networks and trends. 🎨 𝗔𝗿𝘁: Teaches us how to tell stories through engaging data visualizations that resonate with audiences. These 𝗰𝗿𝗼𝘀𝘀-𝗱𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗮𝗿𝘆 perspectives make data science a dynamic, ever-evolving field. 🌟 As a 𝗽𝗵𝘆𝘀𝗶𝗰𝗶𝘀𝘁 turned 𝗱𝗮𝘁𝗮 𝘀𝗰𝗶𝗲𝗻𝘁𝗶𝘀𝘁, I’ve experienced firsthand how skills from different domains can enrich our work. ⚛️ 𝗣𝗵𝘆𝘀𝗶𝗰𝘀 taught me to break down complex problems into simpler components and analyze them systematically. Today, I bring that mindset into every data science project—whether it’s building machine learning models, analyzing large datasets, or creating scalable systems. Combining my analytical approach with insights from other fields has helped me uncover 𝗻𝗲𝘄 𝗽𝗲𝗿𝘀𝗽𝗲𝗰𝘁𝗶𝘃𝗲𝘀 and create 𝗶𝗺𝗽𝗮𝗰𝘁𝗳𝘂𝗹 𝘀𝗼𝗹𝘂𝘁𝗶𝗼𝗻𝘀. 🚀 💡 𝙏𝙝𝙚 𝙞𝙣𝙩𝙚𝙧𝙨𝙚𝙘𝙩𝙞𝙤𝙣 𝙤𝙛 𝙙𝙞𝙨𝙘𝙞𝙥𝙡𝙞𝙣𝙚𝙨 𝙞𝙨 𝙬𝙝𝙚𝙧𝙚 𝙩𝙝𝙚 𝙢𝙖𝙜𝙞𝙘 𝙝𝙖𝙥𝙥𝙚𝙣𝙨. By integrating diverse perspectives, we can solve problems in innovative ways and uncover ideas we’d never have imagined by working within the boundaries of a single field. 🔗 How do you integrate interdisciplinary thinking into your work? What fields have influenced your approach? 💬