Our research cuts across many industry verticals. We enable application-specific and company-specific capabilities and algorithms built on top of fundamental new advances.

Some of our on-going projects are listed below. New projects leverage our existing research strengths and often involve developing new capabilities tailored to the specific challenges facing our industry partners.

Learn more about our approach and what makes us unique in "About Us".


Development of methods based in classical electrostatics for surveying, identifying, and mapping the activity of neuronal networks using multiple extracellular electrodes. The methods that allow the algorithms to sense intracellular and intranetwork dynamics are linear, because the Maxwell equations are linear, but they provide direct information on internal nonlinear processes. This work has applications to engineered (non-biological) complex network system design and control, as well neurobiological applications to the understanding and mitigation of cognitive and neurological disorders.

Advanced statistical data assimilation for functional networks of neurons. These project supports the enhanced use of existing data and contains strategies for finding the global minimum of variational principles. These methods, in theory and in practice, give a clear metric of when they can be used to identify the correct answer - and, equally important, when not. These algorithms also have a wide range of applications to engineered complex network system design and control.

Neuromorphic engineering aimed at reverse engineering the cognitive brain in silicon. This work pioneered learning in silicon systems, with the first silicon integrated circuits to embed learning functionality at the individual transistor level, and has achieved the highest energy efficiency of any computing systems to date at 1 femtojoule of energy per synaptic operation, surpassing the energy efficiency of synaptic transmission in the human brain.

Exploration of the neurobiological basis of complex acoustic signal pattern detection, recognition, classification, and associated cognitive processing. This work integrates large scale extracellular electrophysiological methods in awake behaving animals, intracellular techniques, behavioral control, and computational neuroscience including the adaptation and integration of machine learning techniques mirroring efficient biological operations, in order to target cognitive processes such as attention, learning, and memory.

Robust natural language and semantics acquisition and processing using novel neural computation algorithms. This work is extending existing machine learning methods with the focus being not an understanding or transduction of natural language, but identifying the intent of the language and its semantics. Humans say many different things to mean or imply a common concept, and while natural language as interpreted by humans can capture aspects of intent and context, it is a key area where machines currently fall short and why ‘conversations’ with computers seem artificial and forced. This work is focusing on the development of unique neural-derived algorithms that are robust and produce invariant representations of natural language learning that encode the concept or meaning behind the language. We are also pursuing a similar goal for vision.

Neural computation quantum algorithms for ultra-high parallel data processing and analytics. The computational power of the human brain is ultimately the product of a huge combinatorial process across many spatial and temporal scales spanning many orders of magnitude. Abstractly, this represents a massive solution space that computes, or collapses, to a single solution or state which in turn becomes an input into a higher scale network, ultimately interacting with multiple other inputs in a combinatorial way at the higher scale to produce a single network state for the entire system. There is a natural analogy between the vast combinatorial solution space of the brain’s neural networks and that characterizing quantum computational states that we are exploiting towards the development of new classes of quantum algorithms.

Development of algorithms that will allow autonomous systems to go beyond contextual decision making to contextually relevant and autonomous problem solving. Autonomous systems are typically faced with limited computational resources while at the same time necessitating ever increasing sophisticated local (offline) computations in order to make contextually relevant complex decisions under conditions of limited or selective data. This work is effectively allowing autonomous systems to go one step further in order to ’think’ on their own, achieving contextually relevant creative original ideas and thoughts when presented with a technical problem or challenge the system has not seen before.

Identification of inherently encoded causal dynamic motifs for mining contextual relevance in data. Derivation of algorithms from a generalized theoretical framework that formally describes how in any physically constructible geometric network, the geometry of the network i.e. its physical structure, constrains and bounds signaling and the flow of information through the network, which in turn predicts the emergent dynamics of the network. This work is being applied to big data analytics, pattern identification for cybersecurity, and how signaling and information flows in structural biological neural networks (i.e. the brain’s connectome) give rise to the dynamic function that underlies what the brain can do.

Network systems approach to the pathophysiology of autism spectrum disorder. The theoretical considerations of the previous project can be extended to define a ratio between an effective refractory state of a node (e.g. biological neurons), a period of time which represents the cumulative internal processing time of a node, and the time it takes a signal to reach it from the upstream nodes it is connected to. From this we can derive a set of conditions that define an optimal refraction ratio. Intuitively an optimal ratio results when the effective refractory period of a node is dynamically matched to some tolerance to the time it takes for signals to reach it from upstream connected nodes. This work is testing the hypothesis that there is a mismatch in this ratio in networks of neurons in patients with autism relative to typical healthy networks.