Research

Private Distributed Optimization and Learning

We present a private algorithm for solving distributed machine learning (optimization) problems. We use a non-identifiability based privacy definition. Adversarial participants can only learn an equivalence class of private loss (objective) functions and adversarial observations are compatible with all loss functions in the equivalence class. Our algorithm uses correlated random perturbations along with secure multi party computing type strategies. Under appropriate vertex connectivity conditions on the graph, we can show that our algorithm simultaneously achieves both privacy and accuracy.

  • S. Gade, N. H. Vaidya. “Privacy-Preserving Distributed Learning via Obfuscated Stochastic Gradients”, IEEE Conference on Decision and Control, Miami Beach, FL, 17-19 December, 2018.  (pdf)
  • S. Gade, N. H. Vaidya. “Private Optimization on Networks”, IEEE American Control Conference, Milwaukee, WI, 26-29 June, 2018. (pdf) (link)

Privacy in Computing Nash Equilibrium for Networked Games

We propose a distributed algorithm to compute a Nash equilibrium in aggregate games where players communicate over a fixed undirected network. Our algorithm exploits correlated perturbation to obfuscate information shared over the network. We prove that our algorithm does not reveal private information of players to an honest-but-curious adversary who monitors several nodes in the network. In contrast with differential privacy based algorithms, our method does not sacrifice accuracy of equilibrium computation to provide privacy guarantees.

  • S Gade,  A Winnicki,  S. Bose. “On Privatizing Equilibrium Computation in Aggregate Games over Networks”, IFAC World Congress, Berlin, 2020 (accepted) (preprint).

Distributed Robust Statistics

We are witnessing a resurgence of interest in robust statistical estimators due to their utility in fault-tolerant machine learning. In this work, we propose distributed algorithms for computing robust statistics on data that is corrupted by a Byzantine adversary and stored on several networked computers. Our model is more pertinent for data analytics applications over Internet-of-Things devices, sensor networks and cellular/tablet devices. First, we show that computing robust estimators such as Trimmed Mean, Winsorized Mean, Order Statistics (min, median, max) over distributed data, can be set up as a convex, non-smooth distributed optimization problem. Second, we present a cutting plane based distributed optimization algorithm that converges to the optimum (robust statistic) exponentially fast. We provide convergence and complexity guarantees and discuss some privacy properties of our algorithm and its implications to data privacy.

Privacy – Fault Tolerance trade-off in Distributed Computing

We study the problem of achieving Byzantine Fault Tolerant Consensus and Distributed Optimization under Differential Privacy constraints. Intuitively privacy and fault tolerance are conflicting objectives. Privacy involves adding specialized noise to the response in order to hide private information. Fault tolerance on the other hand involves filtering to eliminate the detrimental impact of faulty nodes. We answer the following fundamental questions:  (a) Can we achieve both privacy and fault tolerance simultaneously? (b) What is the trade-off between privacy with fault tolerance?


Prior Research

Robotic Herding

In this work we develop a technique for controlling and tackling organized bird activity near airports. A novel herding technique, called the n-wavefront herding algorithm is developed. In this boundary control type strategy, the pursuer influences only the robots/birds/nodes on the bounding convex hull (boundary), while ensuring that the robot maintains a safe distance from the flock, avoids flock fragmentation, and improves the cohesiveness in the bird flock. Stability and Performance of the herding strategy are studied. Criteria for exponential stability among bird formations are determined using Graph–Theoretic approaches.

  • S. Gade, A. Paranjape, S-J. Chung. “Robotic Herding using Wavefront Algorithm: Performance and Stability”, AIAA Guidance, Navigation, and Control Conference (SciTech-16), San Diego, CA, 4-8 January, 2016. (pdf) (link)
  • S. Gade, A. Paranjape, S-J. Chung. “Herding a Flock of Birds Approaching an Airport Using an Unmanned Aerial Vehicle”, AIAA Guidance, Navigation, and Control Conference (SciTech-15), Kissimmee, FL, 5-9 January, 2015. (pdf) (link)

Collaborative Missions using Unmanned Aerial Systems

Use of Unmanned Aerial Systems (UAS) in urban search and rescue can provide significant help to the response teams by relaying real time information about the victim location and status, assess damage of the disaster and it can help in active planning of response. An Ant-colony based hybrid search algorithm is presented for applications in search in an urban setting. This algorithm is scalable and robust to loss of agents. Its distributed nature makes it ideal for applications in large scale search operations. Human-in-the-Loop hierarchical control architecture is presented here for UAV flock management. On-line mission reconfiguration strategies are presented and their utility in search effectiveness and efficiency is presented.

  • S. Gade, A. Joshi. “Heterogeneous UAV Swarm System for Target Search in Adversarial Environment”, IEEE International Conference on Control, Communication and Computing (ICCC-13), Trivandrum, India, December 2013. (pdf) (link)
  • S. Gade, A. Joshi. “Human-in-Loop Hierarchical Control of Multi-UAV Systems”, International Conference on Intelligent Unmanned Systems (ICIUS-13), Jaipur, India, September 2013. (pdf)

Vanishing Point Estimation

We propose a algorithm for estimation of vanishing points in structured environments. Our method is non-iterative and presents a closed-form solution. Computational time requirement for 3-P method is shown to be much less than the standard least squares based method.

  • V. Saini, S. Gade, M. Prasad, S. Chatterjee. “The 3-Point Method: A Fast, Accurate and Robust Solution to Vanishing Point Estimation”, International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision (WSCG-13), Pilsen, Czech Republic, June 2013. (pdf) (link)