Traditional machine learning models can be formulated as the {expected risk minimization} (ERM) problem: 
$\min_{\mathbf w\in\mathbb{R}^d}\mathbb E_{\xi}[\ell(\mathbf w; \xi)]$, 
where $\mathbf{w}\in\mathbb{R}^d$ denotes the model parameter, $\xi$ represents training samples, $\ell(\cdot)$ is the loss function. Numerous optimization algorithms, such as stochastic gradient descent (SGD), have been developed to solve the ERM problem. However, a wide range of emerging machine learning models are beyond this class of optimization problems, such as model-agnostic meta-learning. Of particular interest of my research is the \textit{stochastic nested optimization} (\textbf{SNO}) problem, whose objective function has a nested structure. Specifically, I have been focusing on two instances of this kind of problem: \textit{stochastic compositional optimization} (SCO) problems, which cover meta-learning, area-under-the-precision-recall-curve optimization, contrastive self-supervised learning, etc., and \textit{stochastic bilevel optimization} (SBO) problems, which can be applied to meta-learning, hyperparameter optimization, neural network architecture search, etc. 

With the emergence of large-scale distributed data, such as the user data generated on mobile devices or intelligent hardware, it is imperative to develop distributed optimization algorithms for SNO (\textbf{Distributed SNO}). A significant challenge for optimizing distributed SNO problems lies in that the stochastic (hyper-)gradient is a biased estimation of the full gradient. Thus, existing distributed optimization algorithms when applied to them suffer from slow convergence rates. In this talk, I will discuss my recent works about distributed SCO and distributed SBO under both centralized and decentralized settings, including algorithmic details about reducing the bias of stochastic gradient, theoretical convergence rate, and practical machine learning applications, and then highlight challenges for future research.

Distributed Stochastic Nested Optimization for Emerging Machine Learning Models: Algorithm and Theory

Robots will play a crucial role in the future and need to work as a team in increasingly more complex applications. Advances in robotics have laid the hardware foundations for building large-scale multi-robot systems, such as for mobile robots and drones. But how to coordinate robots intelligently is a difficult problem. In this talk, I will introduce intelligent planning algorithms for solving this challenge with a focus on one fundamental problem: letting a large team of agents navigate without collisions in congested environments while minimizing their travel times. I will present principled planning algorithms that can efficiently coordinate hundreds of agents while providing rigorous guarantees on completeness and even optimality. I will also present techniques to apply these algorithms to real-world problems with robustness guarantees, such as warehouse robot coordination, railway planning, and multi-arm assembly.

Intelligent Planning for Large-Scale Multi-Robot Coordination

Today’s investments in aircraft and Naval Aviation Support Equipment (SE) sustainment lack a coordinated response. Current metric sustainment reports address aircraft or SE top degrader equipment separately. To our knowledge there is no current association on the SE degrader investments that will have the greatest impact on aircraft readiness. This research will allow for the transition of these findings into an integrated metric data product report.

Naval Aviation Support Equipment Impact on Aircraft Readiness

Rapidly advancing fields of software technologies and the development of software autonomy frameworks present new opportunities and challenges for the DoD. While these technologies allow the DoD to rapidly augment capabilities, integration complexity increases, which oftentimes overwhelms Autonomy Practitioners, or the human end users of autonomy frameworks. This paper describes the utilization of Research and Autonomy Innovation Development Environment and Repository (RAIDER), a Navy maintained repository of autonomy frameworks, intended to leverage the promise of rapid integration while mitigating integration complexity. In particular, RAIDER encourages coordination of acquisition efforts through the reuse and repurposing of existing products, which ultimately improves system interoperability and frees up resources (e.g. staff, schedule). The acquisition strategy and tools laid out in this paper can be applied to implement new mission capabilities and integrate technologies such as machine learning on existing hardware platforms.

Streamlining Software Acquisition Processes Through the Use of an Autonomy Framework Repository and Environment

Electromagnetics Aircraft Launch System (EMALS) Anomaly Detection

Fast reorientation of space-based sensors is needed to increase ISR capacity and to respond to high-value targeting requests. Rapid reorientation maneuvers are realized by constructing and solving a nonlinear optimization problem that models the physics of the rotational motion as part of the ground process of mission operations. The solution trajectories are then uplinked to the spacecraft for attitude guidance. Attitude guidance commands for a spacecraft are a time-series of incremental rotations that are given to the attitude control system at the closed-loop feedback rate which is typically greater than 10 Hz.

In practical space systems, it is generally not possible to uplink reorientation commands as a finely sampled time-series (at or greater than 10 Hz) since this would quickly exhaust the capacity of a spacecraft’s command buffer. Instead, the current paradigm is to issue a reorientation command as a single time-tagged target orientation. Then, an on-board trajectory generator is used to construct an appropriate rotation command time-series at the closed-loop feedback rate which are then fed to the attitude guidance system. Typically, these on-board trajectory generators do not have the processing power to perform trajectory optimization and instead generate a feasible, but sub-optimal, time-series command set at the appropriate rate.

A potential solution to this problem is to compute and then compress the optimized time-series commands using an autoencoder. An autoencoder is an unsupervised machine learning application of a neural network that conducts a non-linear mapping of the inputs to a lower dimensional space, called the latent space, using the encoder portion of the autoencoder. It then reconstructs the same input from the latent space using the decoder portion of the autoencoder. With such a method, the optimized command sequence can be generated and encoded on the ground, then only the latent space is transmitted to the spacecraft and decoded on-board the spacecraft resulting in the reconstructed optimized rotation sequence at the closed-loop feedback rate. Given the time-series nature of the optimized command sequence, one appropriate architecture for the autoencoder is a long short term memory (LSTM) neural network autoencoder.

In this talk, we demonstrate the methodology for constructing and training the LSTM autoencoder and show the reconstruction error for different degrees of compression represented by decreasing dimensionality of the latent space. We then compare the reconstructed command sequence and the uncompressed time-optimal solutions using a flight simulation software at the closed-loop feedback rate. The result is a new and computationally efficient approach for implementing non-legacy maneuvers on board a spacecraft. There are numerous applications beyond spacecraft where this capability can benefit the DoD such as small unmanned vehicle operations and operations in degraded electromagnetic environments.

A Method to Compress and Transmit Optimal Reorientation Commands to a Spacecraft Using Long Short-Term Memory Neural Network Autoencoders

Human Detection in Drone Images Using YOLO for Search-and-Rescue Operations

In this work  we propose the geologic mapping of the Copernicus crater and the characterization of a potential landing site for a rover mission. Moreover, we have selected 68 scientific targets for in situ analyses and sampling, which can answer the ESA’s scientific questions about the history and geology of the Moon. Copernicus Crater is a relatively young impact complex crater, located on the near side and of about 100km indiameter. Studying Copernicus means having the possibility to get information about lunar crust, lunar interior, cratering process, and lunar explosive volcanism as well as constraining the age of a reference stratigraphic plane of the whole Moon, with an important impact on the dating system applied to the entire inner Solar System. Olivine spectral signature was detected (Liu F. et al., 2011), and this can provide information about the lunar crust and mantle. Absolute-dating Copernicus would improve the lunar chronological curve, used to date surfaces of all the bodies of the inner Solar System, through the crater size frequency distribution. Improving this curve means improving the age estimations of all these bodies. Moreover, this work can assist the foreseen human assisted robotic missions. As an incremental update of the work performed on Tusberti, (2019) we firstly performed a large-scale geological characterization and mapping of Copernicus crater, using remote sensing techniques, analyzing topographic, imaging and spectral data coming from Lunar Reconnaissance Orbiter (LRO), Selenological and Engineering Explorer (SELENE), and Clementine missions where many interesting features were visible (like cinder cones, vents, melt differentiated, fresh craters and presence of olivine). Therefore, we generated two geological maps, one covering the entire Copernicus crater (at a scale of 1:400000) and one focused on its NW quadrant (scale 1:100000). In fact, by analyzing the topography in terms of slope and roughness we infer that this quadrant could be the best candidate for the landing area. Based on those elaborations and on a preliminary hazard study (following the constraints from Pajola M. et al., 2017), we selected a tentative landing ellipse with in mind the engineering constraints of the now deprecated HERACLES mission and the forthcoming EL3 (European Large Lunar Lander). Moreover, the landing ellipse was selected not only for safety reasons but also because 68 potential sampling sites which are located in reachable areas. These sampling sites are carefully selected in order to fulfill the major ESA questions related to Moon exploration. 

Geologic mapping and landing site characterization in Copernicus Crater (Moon)

The work presented evaluates distances in tactile Human-Drone Interaction.
A user study was conducted, in which a miniature-sized drone approached participants from four directions. With future scenarios at home or work in mind, the study was carried out in a seated position. The users were left the choice to engage in physical interaction by stopping the quadrotor by tapping to the quadrotor's frame or to stop it using a provided foot switch to provide more insights about their inhibition threshold.

Underline digital video library

Distributed Stochastic Nested Optimization for Emerging Machine Learning Models: Algorithm and Theory

Intelligent Planning for Large-Scale Multi-Robot Coordination

Naval Aviation Support Equipment Impact on Aircraft Readiness

Streamlining Software Acquisition Processes Through the Use of an Autonomy Framework Repository and Environment

Electromagnetics Aircraft Launch System (EMALS) Anomaly Detection

A Method to Compress and Transmit Optimal Reorientation Commands to a Spacecraft Using Long Short-Term Memory Neural Network Autoencoders

Human Detection in Drone Images Using YOLO for Search-and-Rescue Operations

Geologic mapping and landing site characterization in Copernicus Crater (Moon)

B1: Evaluating Distances in Tactile Human-Drone Interaction

Aperiodic Communication for MPC in Autonomous Cooperative Landing

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES