Few-Shot Learning (FSL) is a machine learning (ML) technique used when there are relatively few training examples available.
ML models work best when they have access to massive datasets. But it’s not always possible to have large reams of data to feed into an ML model.
For example, augmented reality startup TechSee uses machine learning to analyze electronic products for automatic issue detection. With FSL, training their model for a new type of device now requires only a small handful of example images. TechSee just raised a $30M investment round in October.
FSL uses a “meta learning” approach. Which means the machine essentially learns how to learn. So with additional knowledge about similar training tasks, a generalized solution can be formed that helps solve more specific classification problems.
Few-shot is part of the “Machine-Ready Data” meta trend.
1.7MB of data is created every second by every person in the world. But most of this data is not fit for machine consumption. The data is messy, unlabeled, or incomplete.
Unsupervised learning is one solution. It can ingest unlabeled data to train its classification model. Searches for this approach have steadily grown 139% over the last 5 years.
And companies like Appen and Dataiku provide platforms that allow enterprises to transform raw data into AI solutions. In practice, this often means an added “human intelligence” layer between datasets and ML pipelines that perform manual data labelling and data wrangling.
Air fryers have the potential to be the next Instant Pot, which is a product that 20% of US households now own. In fact, a recent survey found that 43.1% of Americans are likely to buy an air fryer in the next 12 months. The same survey also found that consumers are more likely to buy an air fryer than other common kitchen appliances, like toasters, blenders and slow cookers.
Bumped is a fintech app that allows users to get back a small percentage of their purchases in the form of fractional stocks.
According to the company, 85% of consumers prefer stock ownership over traditional loyalty rewards (like points and cash-back).
Bumped recently came out of a 2-year beta period. And the 1.0 launch coincided with a $10.4M Series A in early November, led by Canaan Partners.
Bumped is part of the “Consumer-Facing Payment Tech” meta trend. The Honey browser extension (acquired by PayPal for $4B last year) is probably the best-known product in the space. Other examples of this trend include Ibotta, Acorns, Truebill, and Rakuten.