Cognitive Cybersecurity and the Role of AI

With the rise in scale and complexity of cyberattacks, artificial intelligence is becoming increasingly vital for resource-intensive security measures.

Curated Updates from Life With Data

Life With Data is dedicated to providing distilled machine learning and software engineering updates that highlight significant advancements without unnecessary clutter. This approach allows for frequent, concise insights across the industry, preventing information overload.

The Evolution of Data Infrastructure

This year, we significantly enriched the data available in our reports, collaborating with a wider range of external organizations to refine our findings and strengthen our ties with Stanford HAI. The 2021 report delves into the impact of COVID-19 on AI advancement from various perspectives. Notably, the Technical Performance chapter illustrates how an AI startup leveraged machine learning to expedite drug discovery during the pandemic. Additionally, our Economic chapter reveals that AI hiring and private investments remained robust, with virtual conferences leading to increased participation in AI research discussions.

… continue reading

The RundownArticleFull ReportReport’s Public DataCredit: @indexingaiReverse ETL — A Primer

Emergence of Reverse ETL Solutions

Over the past few years, data infrastructure has undergone remarkable transformation, transitioning from Extract, Transform, Load (ETL) to Extract, Load, Transform (ELT). This new approach allows raw data to be transferred from the source system into a data warehouse or lake before being transformed. Teams are now adopting "reverse ETL," which involves transferring data back into third-party systems for operational use, enhancing the functionality of data.

As teams shift from ETL to ELT, the data warehouse emerges as the definitive source of truth for various data, including customer information dispersed across different platforms. Tools like Fivetran, Airbyte, DBT, Snowflake, and Redshift facilitate this architectural change. Traditionally, data from warehouses was utilized for analytical purposes and business intelligence applications like Looker and Superset. Now, data teams are recognizing the potential for operational analytics, effectively delivering real-time insights where they are most needed within the organization.

… continue reading

The RundownArticleSupersetDBTAirbyteCredit: @astasia on Medium

Collaboration with Facebook on Global Mapping

At Stamen, our expertise lies in cartography and data visualization, enabling clients to effectively communicate complex datasets. For nearly two years, we have collaborated with Facebook engineers, designers, and data specialists to develop a comprehensive global base map utilized by billions of users. This project, which launched in late 2020, represents one of our most ambitious endeavors to date.

… continue reading

The RundownArticleStamenOpenStreetMapCredit: @jonahadkins on Medium

Introducing Model Search: An Open Source Platform

The efficacy of neural networks (NNs) is often contingent on their ability to generalize across various tasks. However, crafting NNs that exhibit strong generalization capabilities poses challenges due to the limited understanding within the research community regarding ideal neural network design.

To tackle these challenges and democratize access to AutoML tools for researchers, we are thrilled to announce the open-source launch of Model Search. This platform assists researchers in efficiently developing optimal ML models. Unlike domain-specific tools, Model Search is domain-agnostic and adaptable, finding the architecture that best aligns with any dataset and problem while minimizing development time and resource consumption. Built on TensorFlow, it can operate on individual machines or within distributed environments.

… continue reading

The RundownArticleGithub google/model_searchBeam SearchCredit: @googleai feat my old classmate @itspviolette!

The Intersection of Machine Learning and Causality

Historically, the domains of machine learning and graphical causality have evolved independently. However, there is a growing interest in leveraging insights from both fields to address significant challenges in modern machine learning research. This includes exploring causal inference concepts and their implications for key issues like transfer and generalization.

A primary focus of research at this intersection is causal representation learning, which seeks to identify high-level causal variables from low-level observations. This exploration highlights the potential for mutual benefits between machine learning and causal analysis.

… continue reading

The RundownArticleCausal Discovery from Heterogeneous/Nonstationary DataCausal Discovery with Continuous Additive Noise ModelsCausal Learning — Bernhard SchölkopfCredit: Paper authors

Understanding AI Through Neuroscience

The quest for true artificial intelligence has long centered on constructing machines that possess cognitive capabilities akin to human thought. However, there is ongoing debate about the extent to which AI should mirror biological intelligence. Early AI efforts were loosely inspired by human decision-making processes, while contemporary deep neural networks draw inspiration from the firing patterns of neurons.

Neuroscientist and entrepreneur Jeff Hawkins aims to bridge the gap between neuroscience and AI, advocating for a deeper understanding of the mechanics of intelligence. After initially pursuing a PhD in neuroscience, Hawkins transitioned to Silicon Valley, ultimately founding Palm Computing and developing the PalmPilot.

… continue reading

The RundownArticleRedwood Center for Theoretical NeuroscienceNumentaA Thousand Brains: A New Theory of IntelligenceCredit: @techreview