Fooling the Markets using Adversarial Perturbations
Targeted universal adversarial perturbations for alpha models in automated trading systems.
- Designed a targeted universal attack that manipulates the input data stream of alpha models.
- Demonstrated transferability across DNN, CNN, and RNN architectures and studied cascading market effects.
Adversarial MLDNN / CNN / RNNQuant research
92% attack success · 0.02% perturbation