Quickly and efficiently building and labeling image datasets for machine learning applications can be a prohibitively time-consuming and expensive activity. SkyScan demonstrates a low-cost system that can capture images of aircraft in flight and automatically label the image with captured metadata.
Improving synthetic data with generative deep learning networks
It is generally thought that a good AI model needs a lot of good data. But what about when it is unfeasible or unreasonable to collect such a large dataset? How best to leverage small datasets for machine learning tasks is an active area of exploration. Synthetic data has the potential to alleviate object rarity and long-tail distributions, provided the synthetic data introduces more signal than noise into the system.
You can now do a lot more with a lot less when it comes to deploying AI systems. Low-power edge inference techniques are fundamentally changing the design of fieldable smart sensors. This project is looking at the evolution of design in AI enabled, acoustic sensors that live at the edge by designing and fielding a smart hydrophone.
Using low-cost sensors to locate the source of an RF signal
These days there are all kinds of invisible signals in the environment around us. If you see a problematic consumer drone in your area, how do you find its operator? In the Birdseye project, we simulated the use of mobile, low-cost sensors to geolocate a moving RF-signal. More specifically, we simulated the collection of RF signals coming off a commercial drone flying overhead to see if we could use it to locate its controller.
IQT Labs is developing a pragmatic, multi-disciplinary approach to auditing AI & ML tools. Our goal is to help people understand the limitations of AI/ML and identify risks before these tools are deployed in high stakes situations. We believe auditing can help diverse stakeholders build trust in emerging technologies...when that trust is warranted. For more info, check out this report which describes our auditing approach and what we found when we audited FakeFinder an Open Source deepfake detection tool.
Batch processing of deepfake detection within videos
Increasingly fake videos are popping up around us because DeepFake generation models are getting better at producing realistic output at an alarming scale. FakeFinder was a hands-on project aimed at better understanding Open Source models that can be effective for debunking such videos at a similar rate.
Verifying the geographic location of outdoor images
Suppose that a photograph has surfaced under dubious circumstances, raising the question of where it was really taken. One potential solution, cross-view image geolocalization (CVIG), is the process of geolocating an outdoor photograph by comparing it to satellite imagery of possible locations. This project touched on each major component of CVIG deep learning.