Hrl Laboratories, LLC

Nicholas A Ketz

Praveen K Pilly

Michael D Howard

Jaehoon Choe

Charles E Martin

Soheil Kolouri

Nigel D Stepp

Matthew E Phillips

Steven W Skorheim

Sasha Strelnikoff

Aashish N Patel

Aashish Patel

Michael Howard

Described is a method for improved performance of agent-based machine learning. The method includes training a reinforcement learning (RL) agent on an image processing task. A dataset of states and corresponding actions is then extracted from the RL agent. A measure of attention is applied to an input space of the RL agent. During action selection by the RL agent, image patches of the input space are extracted based on the applied measure of attention. Portions of a set of inputs are clustered based on similarity to the image patches, generating a set of clusters having cluster centers. Non-semantic concept labels are provided as distances to the cluster centers for each state in the dataset.

System and method for unsupervised concept extraction from reinforcement learning agents

Described is a system for semantically grounded video generation. In operation, the system receives a raw video frame of a scene from one or more sensors on an autonomous platform and then encodes the video into a low-dimensional representation of the scene. The low-dimensional representation is then decoded and fed into a controller model for the autonomous platform. Semantic concepts are extracted and then fed into a world model to predict state and action dynamics of the autonomous platform. A raw observation space is then fed into discriminator networks that operate on frames and videos to determine between real and synthetically generated content. The generative capability of one or more encoders and decoders are modified such that the discriminator networks are unable to distinguish between real and synthetically generated content. Finally, the system recursively generates semantically grounded videos using a conjunction of the world model and controller model.

System and method for semantically grounded video generation

Described is a system for improving generalization of an agent, such as an autonomous vehicle, to unanticipated environmental changes. A set of concepts from the agent's experiences of an environment are extracted and consolidated into an episodic world model. Using the episodic world model, a dream sequence of prospective simulations, based on a selected set of concepts and constrained by the environment's semantics and dynamics, is generated. The dream sequence is converted into a sensor data format, which is used for augmented training of the agent to operate in the environment with improved generalization to unanticipated changes in the environment.

System and method for improved generalization from concept constrained dreams

Described is a system for competency assessment of an autonomous system. The system extracts semantic concepts representing a situation. Actions taken by the autonomous system are associated with semantic concepts that are activated when the actions are taken in the situation. The system measures an outcome of the actions taken in the situation and generates a reward metric. The semantic concepts representing the situation are stored as a memory with the actions taken in the situation and the reward metric as a memory. A prospective simulation is generated based on recall of the memory. A competency metric and an experience metric are determined. Competent operational control of the autonomous system is maintained when at least one of the competency metric and the experience metric is above a minimum value. An alert is generated when at least one of the competency metric and the experience metric falls below the minimum value.

System and method for safety and efficacy override of an autonomous system

A closed-loop system for asynchronous brain control of at least one task includes a brain state decoder configured to decode neural signals of a user into control signals for controlling the at least one task, a task interface module configured to transmit the control signals to the at least one task, store status information including a series of messages regarding each of the at least one task, and select one message of the series of messages regarding the at least one task to transmit to the user, and a brain state encoder configured to map the one message received from the task interface module into brain state montages for transmission to the user.

System and method for asynchronous brain control of one or more tasks

An autonomous navigation system for a vehicle includes a controller configured to control the vehicle, sensors configured to detect objects in a path of the vehicle, nonvolatile memory including an artificial neural network configured to classify the objects detected by the sensors, and a processor. The artificial neural network includes a series of neurons in each of an input layer, at least one hidden layer, and an output layer. The memory includes instructions which, when executed by the processor, cause the processor to train the artificial neural network on a first task, identify, utilizing a contrastive excitation backpropagation algorithm, important neurons for the first task, identify, utilizing a learning algorithm, important synapses between the neurons for the first task based on the important neurons identified, and rigidify the important synapses to achieve selective plasticity of the series of neurons in the artificial neural network.

Artificial neural networks having attention-based selective plasticity and methods of training the same

A method for retraining an artificial neural network trained on data from an old task includes training the artificial neural network on data from a new task different than the old task, calculating a drift, utilizing Sliced Wasserstein Distance, in activation distributions of a series of hidden layer nodes during the training of the artificial neural network with the new task, calculating a number of additional nodes to add to at least one hidden layer based on the drift in the activation distributions, resetting connection weights between input layer nodes, hidden layer nodes, and output layer nodes to values before the training of the artificial neural network on the data from the new task, adding the additional nodes to the at least one hidden layer, and training the artificial neural network on data from the new task.

Artificial neural network and method of training an artificial neural network with epigenetic neurogenesis

A brain-machine interface system configured to decode neural signals to control a target device includes a sensor to sample the neural signals, and a computer-readable storage medium having software instructions, which, when executed by a processor, cause the processor to transform the neural signals into a common representational space stored in the system, provide the common representational space as a state representation to inform an Actor recurrent neural network policy of the system, generate and evaluate, utilizing a deep recurrent neural network of the system having a generative sequence decoder, predictive sequences of control signals, supply a control signal to the target device to achieve an output of the target device, determine an intrinsic biometric-based reward signal, from the common representational space, based on an expectation of the output of the target device, and supply the intrinsic biometric-based reward signal to a Critic model of the system.

System and method for continual decoding of brain states to multi-degree-of-freedom control signals in hands free devices

Described is a system for transcranial stimulation to improve cognitive function. During operation, the system generates a customized stimulation pattern based on damaged white matter. Further, data is obtained representing natural brain oscillations of a subject. Finally, while the subject is awake, one or more electrodes are activated in phase with the natural brain oscillations and based on the customized stimulation pattern.

Transcranial stimulation system and method to improve cognitive function after traumatic brain injury

Described is a system for automated artifact removal to generate a clean signal. During operation, the system selects initial noise components from a multi-channel, pre-processed signal by performing independent component analysis decomposition on the pre-processed signal to separate and rank the independent components as noise components. A clean signal is then generated through optimized selection of the noise components based on a signal quality index in which the noise components are moved from the original pre-processed signal until a sufficient signal quality is received.

Growing community of inventors

Topanga, CA, United States of America

Nicholas A Ketz