Transitioning from big data to collective intelligence.
Today, cloud-based big data AI systems face enormous challenges of privacy, latency, power consumption, storage & computational costs.
Existing Big Data & AI platforms can’t learn simultaneously preserving privacy.
Big Data can’t be uploaded at once, making real-time computing an impossibility or slowness.
Transferring & processing big data in centralized clouds consumes massive amounts of energy.
Utilizing big data centers and huge computation resources leads costly operation and management.
TieSet Inc. offers an intelligence-centric platform called STADLE with continuous, distributed & collaborative learning frameworks that resolves the major problems in data-centric AI systems such as privacy, latency, and high costs of utilizing huge data centers and computation resources. We achieve them by only gathering, maximizing, and sharing intelligence from users.
Our customers significantly benefit from our STADLE platform in many ways to realize AI-empowered applications as we can also enable new business areas where privacy has been a very serious bottleneck, drastic improvement in communication and computation efficiencies is needed, or management & maintenance costs of big data systems need to be reduced.
PRODUCTS & SERVICES
The Next Generation Intelligence-Centric AI Platform
STADLE – Scalable Traceable Adaptive Distributed Learning Platform
AI - data remains at local user devices
with decentralized federated learning - not collecting big data
uploading and downloading AI models while tracking their performance
for more than 10K devices in minutes
edge AI training empowered with federated learning
Want to achieve 100% Data Privacy, Low Latency (1/10000 in transferred data) and No Centralized Power
Privacy-Safe Ad Targeting
Intelligent Dynamic Ad Delivery for the Cookieless Future
TieSet’s patented Decentralized Federated Learning framework enables intelligent ad targeting, without dependence on identifiers like the cookie or mobile device ID. User data stays with the owner, training models on the edge device, and only resulting intelligence is shared.
Consumer sentiment is driving a privacy revolution, with regulations gaining momentum worldwide, and major platforms responding by removing identifiers or requiring users to opt-in to identification. Publishers report earning only 50-70% CPMs on anonymized traffic compared to impressions with a cookie, but by 2022, no major browsers will continue supporting this outdated technology. In the mobile ecosystem, Apple is already moving to automatically opt users out of sharing the IDFA (Identifier for Advertisers), on which app campaign targeting relies.
Such changes in the ecosystem tend to drive more dollars toward the walled gardens, but independent players can compete. Continue driving top CPMs for publishers by delivering the results advertisers expect in a sustainable, ad-supported, privacy-protected internet.
Learning Robotic manipulation from distributed robotic arm clusters using Federated Learning.
Grasping and manipulation in robots require extensive data from the real world of objects with different shapes and sizes and from various environments. In the current approach, robotic arms acquire grasping data in a single facility to train a centralized model. (e.g. Google Robotic Farm) While this approach leads to good results, acquiring data is time consuming and expensive resulting in a non-scalable process for a single company (e.g. Amazon), so this framework does not make general manipulators better. We propose to use Federated Learning to decentralized the learning of such robots.