Welcome to our This Week in AI roundup. Our goal with this roundup is to provide an overview of the week's most important news, papers, and industry developments.
This week we have stories about AI innovations, quantum advances, data roadblocks in AI, and more.
AI Innovations & Quantum Advances
IBMs annual THINK conference started this week with updates on their AI innovations and quantum computing efforts. A few highlights from their press release include:
- The company claims to have made a breakthrough in cloud-based database management:
A breakthrough capability in Cloud Pak for Data that uses AI to help customers get answers to distributed queries as much as 8x faster than previously and at nearly half the cost of other compared data warehouses.
- The "no code" AI paradigm such as Watson Orchestrate continues to gain traction:
Requiring no IT skills to use, Watson Orchestrate enables professionals to initiate work in a very human way, using collaboration tools such as Slack and email in natural language. It also connects to popular business applications like Salesforce, SAP and Workday.
- 120X increase in Qiskit's quantum circuit processing speed thanks to their hybrid-cloud solution:
Together with improvements in both the software and processor performance, this allows Qiskit Runtime to boost the speeds of quantum circuits, the building blocks of quantum algorithms, by 120 times.
Stay up to date with AI
Data Roadblocks for AI
Artificial intelligence has the potential to contribute $15.7 trillion to the global economy by 2030, which is equal to the combined output of China and India.
Instead of spending their time on training models, however, data scientists frequently spend 80% of their time cleaning and handling data. The usage of confidential data to feed machine learning models is prohibited by a number of laws, which presents a data roadblock. Data can be biased as well, as the author writes:
Even if the data is safe and representative of every segment in the population, it can still be unusable because it's incomplete, irrelevant, or out of date.
There is also the issue of low data quality. In order to collect the data that's needed it's possible to create systems to extract data from different sources. One solution to these data roadblocks is generating synthetic data with the same statistical properties as the original:
Generating synthetic data can also provide the scale global organizations need to create the quantity, variety, and granularity of data required so that the resulting models are unbiased, accurate, and complete.
Facebook Teaches AI to Forget
Facebook created an AI technique that helps the model "forget" irrelevant information.
Called Expire-Span, the technique functions by predicting information that is most relevant for the task at hand. It then assigns an expiration date to each piece of data, and the data is discarded after that date has passed. This technique is still in the research stage, their next step is to explore how the underlying technique may be used to integrate various forms of memories into AI systems.
Paper of the Week: MathBERT
In different Natural Language Processing (NLP) tasks, large-scale pre-trained models like BERT and GPT-3 have shown a lot of success. However, adapting them to math-related tasks remains a challenge. The paper proposes MathBERT to understand mathematical formulas and their corresponding contexts.
That's it for this edition of This Week in AI, if you were forwarded this newsletter and would like to receive it you can sign up here.