AGAmita Giri

Amita Giri

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Amita Giri has a total of 25 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • NeuroKinect: A Deep Learning Framework for Hand Kinematics Reconstruction from EEG Signals

    by Sidharth Pancholi, Amita Giri

    Brain-Computer Interface (BCI) technology facilitates direct brain-to-device communication, empowering individuals to control their surroundings through brain signals. However, prevalent BCI methods encounter challenges: a) time-intensive preprocessing, b) improper loss function use, and c) non-intuitive hyperparameter settings. To overcome these challenges, we propose NeuroKinect, a novel deep learning framework designed to reconstruct hand kinematics from electroencephalography (EEG) signals. Trained on Grasp and Lift (GAL) task data with streamlined preprocessing, NeuroKinect enhances computational efficiency. NeuroKinect introduces a task-specific loss function, LStat, which balances correlation and error metrics to better align predicted and actual motion. Driven by scientific intuition, we analyze motor task dynamics via event-related potentials and brain source localization to optimize hyperparameters. Extensive evaluation on EEG data collected during grasp-and-lift tasks demonstrates NeuroKinect’s ability to accurately decode 3D hand movement trajectories. The model achieves strong correlation between predicted and actual motion across all spatial dimensions (mean Pearson coefficients: 0.92±0.015 for X, 0.93±0.019 for Y, and 0.83±0.018 for Z dimensions), supported by low mean squared error values (MSE: 0.016±0.001 for X, 0.015±0.002 for Y, and 0.017±0.005 for Z). These results highlight NeuroKinect’s potential for enhancing real-time BCI systems for motor control and rehabilitation.

  • Open Access

    Localization of Realistic Spatial Patches of Complex Source Activity in MEG and EEG

    by Amita Giri, Lukas Hecker, John C. Mosher, Amir Adler, Dimitrios Pantazis

    Accurate localization of neural sources in Magnetoencephalography (MEG) and Electroencephalography (EEG) is essential for advancing clinical and research applications in neuroscience. Traditional approaches like dipole fitting (e.g., MUSIC, RAP-MUSIC) are limited to discrete focal sources, while distributed source imaging methods (e.g., MNE, sLORETA) assume sources distributed across the cortical surface. These methods, however, often fail to capture sources with complex spatial extents, limiting their accuracy in realistic settings. To address these limitations, we introduce PATCH-AP, an enhanced version of the Alternating Projection (AP) method that effectively localizes both discrete and spatially extended sources. We evaluated PATCH-AP against leading source localization methods, including distributed source imaging techniques (MNE, sLORETA), traditional dipole fitting (AP), and recent extended source methods (Convexity-Champagne (CC), FLEX-AP). PATCH-AP consistently outperformed these methods in simulations, achieving lower Earth Mover’s Distance (EMD) scores—a metric indicating closer alignment with the true source distribution. In tests with real MEG data from a face perception task and auditory task, PATCH-AP demonstrated high alignment with the fusiform face area and auditory cortex region. These results highlight PATCH-AP’s potential to enhance source localization accuracy, promising significant advancements in neuroscience research and clinical diagnostics.


  • Energy-Efficient Hybrid STT-MTJ/CMOS Circuit for Machine Learning-Assisted Neuromorphic Computing Applications

    by Shailendra Yadav, Brajesh Kumar Kaushik, Amita Giri

    The CMOS-based neuromorphic computing system (NCS) face significant challenges, such as increasing energy usage and vast area footprints, surpassing the efficiency of biological brains. Spin transfer torque magnetic tunnel junction (STT-MTJ), a type of spin device@comm offers convenient benefits, including nonvolatility, increased energy efficiency, increased speed of operation, and compatibility with CMOS, making them ideal for energy-efficient spiking NCSs that exhibit neuronal behavior. However, the high energy consumption in spintronic-based NCS, primarily due to the high write current required for MTJ switching, remains a significant challenge as neurons in these systems tend to stay active longer than necessary. To address this challenge, we introduce a novel hybrid STT-MTJ/CMOS write terminate circuit (SM-WTC) that efficiently terminates the MTJ current efficiently after MTJ-state switches, significantly improving energy consumption and speed by 2.6× and 2.3×, compared to conventional NCSs. The proposed SM-WTC technique achieves energy consumption reductions of 52.7%, 58.3%, and 62.18% compared to prior work in real-time sensing (RTS) circuit, common-mode tracking and terminating circuit (CM-TTC), and conventional-NCS, respectively. A Cadence Virtuoso simulation using 65-nm CMOS technology has been used to evaluate the proposed circuit. Furthermore, SM-WTC-based NCS achieves a 67.2% improvement in energy-delay product (EDP) over conventional NCS for image edge detection. These advancements position SM-WTC as a commercially viable solution for next-generation artificial intelligence (AI) accelerators and brain-inspired computing architecture.


  • Open Access

    Localization of Realistic Spatial Patches of Complex Source Activity in MEG

    by Amita Giri, Lukas Hecker, John C. Mosher, Amir Adler, Dimitrios Pantazis

    Accurate localization of neural sources in Magnetoencephalography (MEG) and Electroencephalography (EEG) is essential for advancing clinical and research applications in neuroscience. Traditional approaches like dipole fitting (e.g., MUSIC, RAP-MUSIC) are limited to discrete focal sources, while distributed source imaging methods (e.g., MNE, sLORETA) assume sources distributed across the cortical surface. These methods, however, often fail to capture sources with complex spatial extents, limiting their accuracy in realistic settings. To address these limitations, we introduce PATCH-AP, an enhanced version of the Alternating Projection (AP) method that effectively localizes both discrete and spatially extended sources. We evaluated PATCH-AP against leading source localization methods, including distributed source imaging techniques (MNE, sLORETA), traditional dipole fitting (AP), and recent extended source methods (Convexity-Champagne (CC), FLEX-AP). PATCH-AP consistently outperformed these methods in simulations, achieving lower Earth Mover’s Distance (EMD) scores—a metric indicating closer alignment with the true source distribution. In tests with real MEG data from a face perception task, PATCH-AP demonstrated high alignment with the fusiform face area, a region critical for face processing. These results highlight PATCH-AP’s potential to enhance source localization accuracy, promising significant advancements in neuroscience research and clinical diagnostics.

  • Open Access

    Spatiotemporal Dynamics of Invariant Face Representations in the Human Brain

    by Amita Giri, Grace Smith, Cassia Manting, Katharina Dobs, Amir Adler, Dimitrios Pantazis

    The human brain can effortlessly extract a familiar face’s age, gender, and identity despite dramatic changes in appearance, such as head orientation, lighting, or expression. Yet, the spatiotemporal dynamics underlying this ability, and how they depend on task demands, remain unclear. Here, we used multivariate decoding of magnetoencephalography (MEG) responses and source localization to characterize the emergence of invariant face representations. Human participants viewed natural images of highly familiar celebrities that systematically varied in viewpoint, gender, and age, while performing a one-back task on the identity or the image. Time-resolved decoding revealed that identity information emerged rapidly and became increasingly invariant to viewpoint over time. We observed a temporal hierarchy: view-specific identity information appeared at 64 ms, followed by mirror-invariant representations at 75 ms and fully view-invariant identity at 89 ms. Identity-invariant age and gender information emerged around the same time as view-invariant identity. Task demands modulated only late-stage identity and gender representations, suggesting that early face processing is predominantly feedforward. Source localization at peak decoding times showed consistent involvement of the occipital face area (OFA) and fusiform face area (FFA), with stronger identity and age signals than gender. Our findings reveal the spatiotemporal dynamics by which the brain extracts view-invariant identity from familiar faces, suggest that age and gender are processed in parallel, and show that task demands modulate later processing stages. Together, these results offer new constraints on computational models of face perception.


  • Machine learning and explainable-AI based prediction of gate-all-around ferroelectric-FET: how ML models influence XAI

    by Shailendra Yadav, Nakul Giri, Ashish Maurya, Brajesh Kumar Kaushik, Amita Giri

    A novel integration of machine learning (ML) and eXplainable artificial intelligence (XAI) based prediction is proposed to investigate the variability of nanowire (NW) gate-all-around (GAA) ferroelectric-field effect transistors (Fe-FETs). XAI methods such as local interpretable model-agnostic explanations (LIME) and shapley additive explanations (SHAP) enhance the explainability and robustness of ML algorithms for end-users. The NW-GAA-ferro-FETs show tremendous potential for neuromorphic computing systems and compatibility with complementary-metal-oxide-semiconductor technology. The GAA-ferro-FET model is validated using sentaurus technology computer-aided design simulations and experimental data. In this work, the first-ever ML algorithms for NW-GAA-ferro-FETs are proposed, achieving physics-based TCAD accuracy with faster learning and lower computational cost. Compared to ML-based algorithms, physics-based simulation of conventional emerging devices requires a high level of device information and a substantial amount of time to provide correct findings and well-fit models. The ML algorithm achieved a R2-score of 99.96%, a lower mean square error, and completed the average inference in just 71.82 milliseconds, compared to TCAD simulations that would take 400 h (=17 days) to process 5000 samples. The results indicate that the novel integration of ML and XAI can lead to a substantial reduction in the computational cost associated with various emerging FET devices, such as ferro-FET, feedback FET, tunnel FET, 2D material-based FET, spin-FET, bio-FET, and other next-generation FETs. End-users can receive suggestions and warnings about potential errors before initiating the investigation process, this helps speed up the development of ferro-FET and other next-generation FETs for use in aerospace, defence, and space exploration.


  • Flexible Alternating Projection for Spatially Extended Brain Source Localization

    by Lukas Hecker, Amita Giri, Dimitrios Pantazis, Amir Adler

    Objective: Understanding neural sources behind MEG and EEG signals is significant for basic and clinical neuroscience. Existing techniques, such as Recursively Applied and Projected Multiple Signal Classification (RAP-MUSIC) and Alternating Projection (AP), rely on limited current dipoles, representing focal sources with zero spatial extent. However, this oversimplifies realistic neural activity, which exhibits varying spatial extents. Methods: To address this, we enhanced the AP approach, creating FLEX-AP, capable of localizing discrete and extended sources. FLEX-AP simultaneously optimizes location and spatial extent of candidate sources. Results: FLEX-AP demonstrated superior performance, reducing localization errors compared to AP, RAP-MUSIC, and FLEX-RAP-MUSIC, particularly with extended sources. Moreover, FLEX-AP exhibited enhanced robustness against modeling errors in realistic scenarios. Applying FLEX-AP to MEG recordings of auditory responses validated its effectiveness, underscoring potential in advancing neuroscientific investigations. Conclusion: FLEX-AP offers a robust, flexible framework for M/EEG source localization, overcoming limitations of simplistic zero-extent dipole models. By accurately estimating position and spatial extent of neural sources, FLEX-AP bridges the gap between theoretical models and realistic activity, demonstrating utility in simulated and real-world scenarios. Significance: FLEX-AP advances source imaging techniques with implications for basic neuroscience and clinical applications. Modeling extended sources precisely enables more accurate investigations of brain dynamics, potentially improving diagnostic and therapeutic approaches for neurological and psychiatric disorders.


  • Open Access

    Localization of Spatially Extended Brain Sources by Flexible Alternating Projection (Flex-AP)

    by Lukas Hecker, Amita Giri, Dimitrios Pantazis, Amir Adler

    Magnetoencephalography (MEG) and electroencephalography (EEG) are widely employed techniques for the in-vivo measurement of neural activity with exceptional temporal resolution. Modeling the neural sources underlying these signals is of high interest for both neuroscience research and pathology. The method of Alternating Projection (AP) was recently shown to outperform the well-established recursively applied and projected multiple signal classification (RAP-MUSIC) algorithm. In this work, we further enhanced AP to allow for source extent estimation, a novel approach termed flexible extent AP (FLEX-AP). We found that FLEX-AP achieves significantly lower errors for spatially coherent sources compared to AP, RAP-MUSIC, and the corresponding extension, FLEX-RAP-MUSIC. We also found an advantage for discrete dipoles under forward modeling errors encountered in real-world scenarios. Together, our results indicate that the FLEX-AP method can unify dipole fitting and distributed source imaging into a single algorithm with promising accuracy.

  • Open Access

    An F-ratio-based method for estimating the number of active sources in MEG

    by Amita Giri, John C. Mosher, Amir Adler, Dimitrios Pantazis

    Introduction: Magnetoencephalography (MEG) is a powerful technique for studying the human brain function. However, accurately estimating the number of sources that contribute to the MEG recordings remains a challenging problem due to the low signal-to-noise ratio (SNR), the presence of correlated sources, inaccuracies in head modeling, and variations in individual anatomy.Methods: To address these issues, our study introduces a robust method for accurately estimating the number of active sources in the brain based on the F-ratio statistical approach, which allows for a comparison between a full model with a higher number of sources and a reduced model with fewer sources. Using this approach, we developed a formal statistical procedure that sequentially increases the number of sources in the multiple dipole localization problem until all sources are found.Results: Our results revealed that the selection of thresholds plays a critical role in determining the method's overall performance, and appropriate thresholds needed to be adjusted for the number of sources and SNR levels, while they remained largely invariant to different inter-source correlations, translational modeling inaccuracies, and different cortical anatomies. By identifying optimal thresholds and validating our F-ratio-based method in simulated, real phantom, and human MEG data, we demonstrated the superiority of our F-ratio-based method over existing state-of-the-art statistical approaches, such as the Akaike Information Criterion (AIC) and Minimum Description Length (MDL).Discussion: Overall, when tuned for optimal selection of thresholds, our method offers researchers a precise tool to estimate the true number of active brain sources and accurately model brain function.


  • Open Access

    An F-ratio-Based Method for Estimating the Number of Active Sources in MEG

    by Amita Giri, John C. Mosher, Amir Adler, Dimitrios Pantazis

    Magnetoencephalography (MEG) is a powerful technique for studying the human brain function. However, accurately estimating the number of sources that contribute to the MEG recordings remains a challenging problem due to the low signal-to-noise ratio (SNR), the presence of correlated sources, inaccuracies in head modeling, and variations in individual anatomy. To address these issues, our study introduces a robust method for accurately estimating the number of active sources in the brain based on the F-ratio statistical approach, which allows for a comparison between a full model with a higher number of sources and a reduced model with fewer sources. Using this approach, we developed a formal statistical procedure that sequentially increases the number of sources in the multiple dipole localization problem until all sources are found. Our results revealed that the selection of thresholds plays a critical role in determining the method`s overall performance, and appropriate thresholds needed to be adjusted for the number of sources and SNR levels, while they remained largely invariant to different inter-source correlations, modeling inaccuracies, and different cortical anatomies. By identifying optimal thresholds and validating our F-ratio-based method in simulated, real phantom, and human MEG data, we demonstrated the superiority of our F-ratio-based method over existing state-of-the-art statistical approaches, such as the Akaike Information Criterion (AIC) and Minimum Description Length (MDL). Overall, when tuned for optimal selection of thresholds, our method offers researchers a precise tool to estimate the true number of active brain sources and accurately model brain function.