On a Mission to Democratize AI

The first is through our Labeling Provider Marketplace, which helps small and medium-size labeling companies find customers and advertise their work. All labeling partners on our platform have their own social mission: some employ people with special needs, others focus on the disadvantaged communities that they serve. We also provide them with data-driven, customized feedback through our Labeling Partner Portal which allows them to provide the proper training to their employees and increase their earning potential.

The second way is through our fight against the myth that Big Data is king in Machine Learning; we jokingly call ourselves the “Data Slayers”, which is also the name of our online community. Our B2B platform empowers companies of all sizes to reduce the amount of data that they use to train models, dramatically improving their margins. Meanwhile, the community version of the platform serves two purposes: on the one hand, it helps propagate the message that building high-performance models with less data is possible (in fact, Data Curation often helps get better results), and it educates the next generation of ML scientists on how to select data efficiently – both relying on friendly competitions through which participants can earn swag, free ML books and online courses.

At Alectio, we strive to provide solutions that not only protects our customer’s privacy, but actually increases it. That’s why we decided to use an approach inspired by Federated Learning when developing our SDK to let our users benefit from our Data Curation technology without exposing their model or their data. Even better: by curating their data, our users eventually need to expose only fractions of their datasets to third-party companies or can avoid third-parties altogether when the compression is high enough that they can annotate data internally, preserving their and their customers’ data privacy.

At Alectio, we think about this constantly. Our goal is to dramatically reduce the amount of data you need for your machine learning projects so that AI is more accessible, less expensive, and more environmentally sustainable. We work with ethical labeling partners who care about their employees as much as they care about the quality of their work.

Responsible AI is another important topic for our team. We’re constantly thinking about ways to leverage our technology as an explainability tool, to provide transparency regarding the data records that caused a specific behavior in the model. While traditional explainability frameworks can only provide a reason why the model made a specific decision (i.e., a person is denied a loan because of their zip code), ours can state which records taught the model to flag people based on the location where they live, giving an opportunity to data scientists to correct biases instead of just identifying them. We have A major breakthrough in our research was the discovery that because most facial recognition models struggle to interpret the faces of dark-skinned people due to the difficult to identify shadows, a fair facial recognition algorithm would require a training dataset with more dark-skinned faces than white ones, challenging the traditional beliefs that classes should be fully balanced.

Alectio’s core technology is based on reducing the amount of data needed to train a Machine Learning model. It means that with this technology, data scientists won’t just spend less time and money training models, but will also reduce their carbon footprint. That certainly won’t be enough to save the planet, but it is a very important step in the right direction.

Synthetic Data Generation is another topic that gives us concern at Alectio. We believe in its positive impact on AI research, but the Generative Adversarial Networks (a.k.a, GANs) used to generate data are among the most energy-inefficient models. In order to overcome this, we are adapting our curation technology to guide a synthetic data generation process into generating data strategically, so that less data needs to be created to reach similar results.

That being said, we know that companies will keep working with natural data for years to come, and had to make a difference there, too. Today, we’re moving the needle there by helping companies, in particular in the autonomous vehicles space, develop smart data collection strategies to avoid driving aimlessly in circumstances where informative data is sparse.