At UnifyID labs, we strive to establish and maintain an innovation-first approach. We imbibe a pragmatic, egalitarian and anti-dogmatic approach when it comes to data modeling.
On this page, you will find resources pertaining to our research including peer-reviewed publications (yes, we are a startup that publishes papers!), technical essays and datasets.
Our research spans several areas:
Tensor compression based pre-processing of sensor inputs that includes studying the compression invariant behavior of CNNs to CPD compressed tensor inputs
Decision fusion frameworks
Time series analysis, especially in the irregularly sampled regime
Adversarial attacks and defenses for deep neural networks
Initialization strategies for training compressed neural networks
CNN deploying on off-cloud computation-constrained environs such as smart-phones and embedded systems
RNN compression strategies
Analyzing generalization behavior of CNNs
Implicit deep generative models
Supervised dataset collation
Physics-inspired data modeling
In case you are intrigued by our story and would like to join our ranks, please apply. Frequentists, Bayesians, deep learners, shallow learners, Geometers, Topologists, Non-parametricians, Heavy parametricians, Tensor-flowers, PyTorchers are all welcome!
CPD: Canonical Polyadic Decomposition
CNN: Convolutional Neural Network
RNN: Recurrent Neural Network
ML: Machine Learning
UnifyID has presented at workshops and talks at the following conferences: ICML 2017 in Sydney, CVPR 2017 in Honolulu, and NIPS 2016 in Barcelona.
Smile in the face of adversity much? A print based spoofing attack
In this paper, we demonstrate a simple face spoof attack targeting the face recognition system of a widely available commercial smart-phone.
Vulnerability of deep learning-based gait biometric recognition to adversarial perturbations
In this paper, we would like to draw attention towards the vulnerability of the motion sensor-based gait biometric in deep learning-based implicit authentication solutions.