Panelist Abstracts

Topic 1: Epilepsy & Cortical Disorders (Neuromodulation)

Session 1 Abstract Submissions/Responses

    • Giacomo Valle

      Abstract: What are the "Needs, Challenges, Contributions, Impact and Investment" opportunities for the field of neuromorphic technologies in medicine?

    • Kendall Lee

      Abstract: Modern neural engineering technologies now allow for control of neurotransmitter levels by deep brain stimulation. Artificial Intelligence will allow for automated neural control algorithms in the future.

      Further complex electro-analytical methodologies such as MCSWV, will also be available for neuropsychiatric therapies.

    • Chris Rozell

      Abstract: Currently, commercial neurotechnologies are designed with very minimal processing capability, geared almost entirely toward very specific notions of what onboard processing is necessary for closed loop stimulation to treat specific disorders such as epilepsy and Parkinson's disease. There are presumably many constraints driving this narrow product design, including power consumption, form factor limitations, and the time necessary for new engineering designs that meet regulatory standards. Future innovations would benefit from an expanded set of capabilities in a package that improves the user experience in ways that will increase accessibility (e.g., smaller form factors, longer batter life, etc.).

    • Sydney Cash

      Abstract: Neural interfaces, neuromodulatory systems and brain-computer interfaces, have made significant strides over the past decade, evidenced by ongoing clinical trials and increasing clinical applications. However, substantial challenges remain in harnessing their full potential for clinical use. On the engineering side, challenges such as energy efficiency, packaging, data management, and security are identified and the focus of research in multiple groups.

Topic 2: Human-machine interfaces (Prosthetics)

Session 2 Abstract Submissions/Responses

    • R. James Cotton

      Title: Clinical Gaps and Understanding Real World Outcomes with AI Technologies

      Abstract: I will first discuss important clinical gaps for both upper and lower prosthesis users given current the existing technologies, touching on performance, safety, and usability. I will then emphasize the importance of understanding and optimizing real world function and the risks of an over-emphasize on laboratory measures and clinical outcomes assessments as surrogates for real world function. Finally, I will discuss how advances in computer vision and wearable sensors can better link these constructs and inform research.

    • Ranu Jung

      Abstract: Neurotechnology holds the promise of enhancing our understanding and offering innovative treatment of both physical and mental health conditions. However, transitioning these advancements from the laboratory to clinical and home settings to transform healthcare remains a significant challenge. This talk discusses the use of a convergence approach for pioneering advancements in neurotechnology to revolutionize healthcare. The journey from concept to development and deployment of a Neural-Enabled Prosthetic Hand System in an FDA approved first-in-human trial will be used to illustrate the extensive collaboration across government, industry, academia, and the volunteer participants necessary to make major advances in human-machine interfaces.

    • Elisa Donati

      Title: Neuromorphic Hardware for closed-loop in Healthcare

      Abstract: Neuromorphic technology is at the forefront of revolutionizing healthcare by enabling the development of efficient, bio-inspired closed-loop systems that directly interact with biological tissues in real-time. In this talk, I will delve into how neuromorphic devices, which mimic the neural architecture of the brain, are transforming medical technologies, particularly in the areas of neuroprosthetics, wearable monitors, and chronic disease management.

      A key focus will be on how neuromorphic chips can replicate the behavior of biological neurons with high precision, facilitating seamless bioelectric signal processing for real-time, adaptive feedback. This is critical for closed-loop systems, where continuous monitoring and immediate adjustments to external stimuli are necessary to provide tailored therapeutic interventions. By leveraging these real-time capabilities, neuromorphic technologies allow for more autonomous, responsive medical devices that can improve patient outcomes, reduce intervention needs, and enhance the overall quality of life.

    • Luke Osborn

      Abstract: Luke Osborn: Neuromorphic encoding is necessary for reducing complex, high dimensional sensor signals (e.g., tactile, vision, auditory) into meaningful features for driving useful sensory feedback stimulation paradigms.

    • Xing Chen

      Abstract: Needs: Clinical solutions that are safe, long-lived and effective, providing assistance in daily tasks, object recognition, navigation, and independence. Challenges: Technological challenges in interfacing with the brain with sufficient resolution and specificity, allowing neuromodulation without inducing seizures and aberrant activity, building devices that are easy to implant across clinical sites and individual patients. Contributions: Development of biocompatible materials, scientific understanding of neural encoding and decoding to improve specificity and efficacy of neuromodulation techniques. Impact: Providing hope and inspiration for blind and visually impaired people, and laying the foundation for breakthrough devices. Investment: Harnessing solutions from the lab and bringing them to patients. Recruitment of patients and other stakeholders to shape and guide the field from an early stage. Raising awareness amongst the general public of both the promise and pitfalls of this technology.

    • Helen Huang

      Abstract: 

      1. The design principles for neural-machine interfaces (NMIs) should be anchored in established human motor control and learning theories.
      2. The design of an efferent neural decoder should be guided by human neuromechanics.
      3. The design of an afferent neural encoder should aim to enhance sensorimotor integration
      4. Neuromorphic computing and artificial intelligence are powerful tools for addressing NMI adaptations to individual users and for evolving over time, with low power consumption and distributed processing.

    • Nitish Thakor

      Abstract: Prostheses, whether upper or lower limb for amputees, have concentrated on achieving mechatronic motion and control. Control is achieved through muscle, nerve or direct cortical interface, and using either noninvasive or invasive approaches. However, prosthesis in general and neuroprosthesis in particular are only now attending to incorporating sensory encoding and feedback. I will discuss the need to develop different sensory modalities and mimicking the sensory modalities in biomimetic manner. This would involve sensors that mimic receptors in the skin and encode them as neural spiking activity. This approach has many advantages in terms of encoding, but also poses challenges in terms of novel algorithms and novel approaches to neural interface and feedback. Therefore, my goal is to present some ideas and provoke discussion on sensors, sensory encoding, neural interface and achieving sensory prosthesis solutions for building future closed loop brain machine interface technologies.

    • Shih-Chii Liu

      Abstract: Needs: How to develop a closed-loop system that can dynamically analyze the underlying responses of the area being stimulated for a prosthesis and to provide continuous meaningful feedback to the network outputting the stimulation signals.

      Challenges: On the modelling side, how can we build a model that can extract the needed information from the responses of the neurons in the stimulation area and also deals with the general variability of their responses over days? And how to build the decoder that generates the feedback signal from the currently relatively small number of stimulation electrodes?

      Contribution from Neuromorphic: For the technology and computing side, how can we ensure that the system can analyze the recordings in real-time and provide sparse, low-latency meaningful feedback signals with similar time constants as the brain area under stimulation. Neuromorphic sensors are already e.g. the neuromorphic event camera DVS produces an event-triggered low-latency stream of spikes to transient changes in the scene, for e.g. the DVS is used as the camera in the work by Botond Roska and partners to restore vision using optogenetics.”

Topic 3: Materials for Neuromorphics (Devices)

Session 3 Abstract Submissions/Responses

    • Duygu Kuzum

      Abstract: The next leap in implantable neural interfaces requires technological advances in materials, devices, and computing paradigms. Multimodal approaches integrating optical and electrical sensing modalities can overcome spatiotemporal resolution limits of neural sensing as well as open up new avenues for non-invasive neural recording. Integration of sensing, computation and memory on a single array can enable real-time processing of neural signals for compact, low-power and high-throughput neuromorphic brain machine interfaces. Here, I will present this vision, its challenges, and discuss recent advances in the areas of transparent neural interfaces for multimodal recordings, neuromorphic approaches for on-chip neural processing and computational co-design at the system level for minimally invasive neural interfaces.

    • M. Anthony Lewis

      Abstract: M. Anthony Lewis: Intelligent devices, from wearable health monitors to implants, increasingly surround us. As these technologies evolve, more efficient AI systems are crucial. Traditional deep learning, while powerful, struggles with edge deployment due to high computational and energy demands. We introduce Flow Machines, a novel paradigm for biomedical wearables leveraging state space models and neuromorphic computing. Unlike conventional frame-based deep learning, Flow Machines create internal representations of the external world incrementally, utilizing processing history. This approach contrasts with typical edge-deployed deep learning networks that recompute from scratch, ignoring past computations. Flow Machines accumulate internal representations over time, enabling more compact and efficient AI models suitable for resource-constrained wearables. By transcending traditional paradigms, we aim to enable sophisticated analysis and decision-making in a new class of intelligent biomedical devices. Our research co-optimizes Flow Machines using state space models as their core architecture. This method addresses edge computing limitations by leveraging internal state and incremental learning, achieving state-of-the-art performance across various edge challenges, from health monitoring to adaptive therapeutics. Flow Machines promise to revolutionize wearable healthcare technology, paving the way for personalized, context aware solutions and improved patient outcomes.

    • Gert Cauwenberghs

      Abstract: Our work offers an efficient neuromorphic hardware-software co-design framework rectifying the energy-efficiency imbalance between the training and the inference phases in machine learning (ML) artificial intelligence (AI) systems. It leverages the great tolerance of task performance achievable during inference to memory leakage during training, to implement neuromorphic paradigms for energy-efficient memory throughout the learning phase. Hence it substantially reduces the overall energy footprint of ML-based AI, as it is dominated by the energy costs of training which requires significantly more parameter updates and memory writes compared to inference. Operating at multiple time-scales to implement synaptic metaplasticity where fast episodic-memory work in conjunction with one-shot continual online learning architectures, these learning-in memory systems strike a balance between the adaptation rate of the synaptic elements and their parameter retention capability to realize an optimal traded-off with respect to the energy dissipated per memory write.

    • Shantanu Chakrabarty

      Abstract: Hybrid neural interfaces, comprising biological and synthetic neurons, provide a pathway for designing the next generation of neuroengineering systems. Existing wetware hardware implementations have non-overlapping architectural and operational boundaries, with each subsystem optimized independently using platform-specific signal representations. In contrast, a true hybrid neural interface should facilitate cross-mapping between biological and synthetic neurons, enabling the system to seamlessly optimize and delegate computational capabilities between the two domains. This type of interface would not only better control the capabilities of biological systems at the neural circuit level but also augment and repair neural functions at the system level. 

    • Jennifer Hasler

      Title: What Steps are Required to Reach a Silicon Cortex? 

      Abstract: We are now over 10 years from the Neurmorphic Roadmap paper outlining a path towards building a silicon cortex, and over 15 years from the start of the DARPA SyNAPSE program that promised to build a mouse brain at the end of the end of its five year program. A number of electronics advances over these last several years, as well as the wide interest in Neural Networks / Neuromorphic Engineering, make the opportunities even more possible if we understand the next major steps along these directions. These directions can have profound impact on both engineering applications and neuroscience. This discussion will highlight a few of these questions, including the importance of analog computation, neuron + dendritic models, and the need to effectively model the engineering-level computation and build cortical columns and cortical layers.

    • Ulkuhan Guler

      Abstract: Biomedical sensors generate continuous streams of data, often requiring real time processing to ensure timely, accurate analysis and prompt action based on the interpretation of this information. By implementing neuromorphic techniques directly on chip, data can be processed efficiently with low power consumption, enabling wireless transmission of the processed signals. This reduces the need for off-chip data handling, resulting in compact, energy-efficient, and highly responsive biomedical systems, which are crucial for wearable or implantable health monitoring devices.

Topic 4: (Medical Imaging, wearables and analysis)

Session 4 Abstract Submissions/Responses

    • Amparo Guemes

      Abstract: Peripheral nerve recording and stimulation are vital for advancing neuromodulation therapies for chronic conditions like chronic pain, epilepsy, and diabetes. Conventional neural interfaces are limited by high power consumption, low adaptability, and the complexity of data processing in long-term applications. The demand for implantable devices that interface with the nervous system in real time necessitates innovative solutions. Neuromorphic computing utilizes biologically inspired designs characterized by event-based, asynchronous processing, which enables efficient real-time data handling while significantly reducing energy consumption, making it a promising approach. Spiking Neural Networks (SNNs) in this field allow for more biologically plausible representations of neural activity, enabling precise classification of neural signals and improved decision making for stimulation protocols. Moreover, SNNs can adaptively modify their response based on incoming signals, allowing for real-time adjustments in stimulation protocols. 

    • Pamela Abshire

      Abstract: The truly surprising thing about recent advances in neural networks (i.e. large language model etc.) is the extent to which functions that we generally think of as cognitive – like language and writing – can be accomplished by what is essentially the computational equivalent of a lookup table. What these neural networks are actually doing is finding the patterns and correlations inherent in these problems, and extrapolating from them to produce surprisingly complex outputs that seem to express thoughts. What they are really doing is piecing together bits of patterns from huge libraries of inputs in these spaces that they have seen many many many times, until they are able to pick out the patterns in these extremely high dimensional spaces.

    • Robert Stevens

      Abstract: Challenges Acute brain injuries (ABI) namely those sustained in the setting of trauma or cerebrovascular
      disease are leading global causes of death and disability, and they increase the likelihood of later-onset neurodegenerative disorders. Current interventions to treat ABI are marginally effective or ineffective, perhaps owing to unpredictable treatment effects seen in heterogeneous
      patient samples

  • Registration