27 research outputs found
Using Mixture of Experts to Fine-Tune Robotic Video Transformers
No full textVideo generation models have recently showcased impressive results, generating high quality visual features with realistic physics and motion. Such video generators are intriguing for robotics because after fine-tuning to the robotic embodiment, have the potential to serve as generalizable world models and real-world simulators. Among video generation approaches, masked video transformers provide a computationally efficient alternative to diffusion-based methods. Building on recent successes of Mixture of Experts (MoE) in transformer architectures, we propose a novel approach to improve pre-trained robotic video transformers using sparsely gated MoE. Our method replaces the feedforward layers of the transformer block with sparely gated MoE layers. We also introduce an innovative weight initialization scheme that improves training convergence while fine-tuning masked video transformers. We evaluate our method on the 1xgpt humanoid robotic dataset, demonstrating improvements in both cross-entropy loss (0.07 reduction) and LPIPS scores (0.007 reduction). Our findings suggest that MoE-based fine-tuning with strategic weight initialization can enhance the performance of robotic video transformers while maintaining computational efficiency through sparse expert activation
An ECG Processor for the Detection of Eight Cardiac Arrhythmias with Minimum False Alarms
No full textInfluence of stearic acid and alumina nanofluid on CO2 wettability of calcite substrates: Implications for CO2 geological storage in carbonate reservoirs
No full textHypothesis: Atmospheric CO2 emissions trigger global warming and climate change challenges. Thus, geological CO2 storage appears to be the most viable choice to mitigate CO2 emissions in the atmosphere. However, the adsorption capacity of reservoir rock in the presence of diverse geological conditions, including organic acids, temperature, and pressure, can cause reduced certainty for CO2 storage and injection problems. Wettability is critical in measuring the adsorption behavior of rock in various reservoir fluids and conditions.
Experiment: We systematically evaluated the CO2-wettability of calcite substrates at geological conditions (323 K and 0.1, 10, and 25 MPa) in the presence of stearic acid (a replicate realistic reservoir organic material contamination). Similarly, to reverse the effects of organics on wettability, we treated calcite substrates with various alumina nanofluid concentrations (0.05, 0.1, 0.25, and 0.75 wt%) and evaluated the CO2-wettability of calcite substrates at similar geological conditions.
Findings: Stearic acid profoundly affects the contact angle of calcite substrates where wettability shifts from intermediate to CO2-wet conditions, reducing the CO2 geological storage potential. The treatment of organic acid-aged calcite substrates with alumina nanofluid reversed the wettability to a more hydrophilic state, increasing CO2 storage certainty. Further, the optimum concentration displaying the optimum potential for changing the wettability in organic acid-aged calcite substrates was 0.25 wt%. The effect of organics and nanofluids should be augmented to improve the feasibility of CO2 geological projects at the industrial scale for reduced containment security.The first and second authors acknowledge the Australian Government’s scholarship (Research Training Program) for their higher studies and Curtin University for supervision and resources
An 8.7 mu J/class. FFT accelerator and DNN-based configurable SoC for Multi-Class Chronic Neurological Disorder Detection
No full textIN
A Low Power Multi-Class Migraine Detection Processor Based on Somatosensory Evoked Potentials
No full textMigraine is a disabling neurological disorder that can be recurrent and persist for long durations. The continuous monitoring of the brain activities can enable the patient to respond on time before the occurrence of the approaching migraine episode to minimize the severity. Therefore, there is a need for a wearable device that can ensure the early diagnosis of a migraine attack. This brief presents a low latency, and power-efficient feature extraction and classification processor for the early detection of a migraine attack. Somatosensory Evoked Potentials (SEP) are utilized to monitor the migraine patterns in an ambulatory environment aiming to have a processor integrated on-sensor for power-efficient and timely intervention. In this work, a complete digital design of the wearable environment is proposed. It allows the extraction of multiple features including multiple power spectral bands using 256-point fast Fourier transform (FFT), root mean square of late HFO bursts and latency of N20 peak. These features are then classified using a multi-classification artificial neural network (ANN)-based classifier which is also realized on the chip. The proposed processor is placed and routed in a 180nm CMOS with an active area of 0.5mm(2). The total power consumption is 249 mu W while operating at a 20MHz clock with full computations completed in 1.31ms.IN
Characterizing Accuracy Trade-offs of EEG Applications on Embedded HMPs
No full textElectroencephalography (EEG) recordings are analyzed using battery-powered
wearable devices to monitor brain activities and neurological disorders. These
applications require long and continuous processing to generate feasible
results. However, wearable devices are constrained with limited energy and
computation resources, owing to their small sizes for practical use cases.
Embedded heterogeneous multi-core platforms (HMPs) can provide better
performance within limited energy budgets for EEG applications. Error
resilience of the EEG application pipeline can be exploited further to maximize
the performance and energy gains with HMPs. However, disciplined tuning of
approximation on embedded HMPs requires a thorough exploration of the
accuracy-performance-power trade-off space. In this work, we characterize the
error resilience of three EEG applications, including Epileptic Seizure
Detection, Sleep Stage Classification, and Stress Detection on the real-world
embedded HMP test-bed of the Odroid XU3 platform. We present a combinatorial
evaluation of power-performance-accuracy trade-offs of EEG applications at
different approximation, power, and performance levels to provide insights into
the disciplined tuning of approximation in EEG applications on embedded
platforms.Comment: 7 pages, 10 figure
Author Correction: Bioinformatics approach for the construction of multiple epitope vaccine against omicron variant of SARS-CoV-2
No full textPediatric brainstem gliomas: An institutional experience
No full textObjective: The aim of this study was to analyze the clinical profiles and outcomes of pediatric brainstem gliomas treated at our institute.Methodology: We reviewed the files of 18 pediatric age group patients diagnosed with brainstem glioma at our institution. The following variables were recorded: age, sex, duration of symptoms, date of diagnosis, main clinical symptoms, Karnofsky performance status score, magnetic resonance imaging findings, histopathology findings, details of the treatment given, disease progression, and date of mortality/last follow-up. This data were then transferred to SPSS version 23 which was used for further analysis.Results: The mean age of our cohort was 8.6 years (range 3-15). There were 11 (61.1%) males and 7 (38.9%) females. There were 16 (88.9%) patients with diffuse intrinsic pontine gliomas (DIPGs), 1 (5.6%) patients with exophytic medullary gliomas, and 1 (5.6%) patient with midbrain/tectal glioma. Mean overall survival (OS) was 9.7 months. Mean progression-free survival (PFS) was 6.3 months. All patients with DIPG eventually passed away from their disease. Patients with DIPG who received radiotherapy had a longer OS and PFS than those who did not (9.8 and 6 months vs. 3.4 and 2.4 months). Diagnostic latency \u3e1 month was found to have a statistically significant longer progression-free interval.Conclusion: DIPGs in the pediatric population have a poor prognosis. Radiotherapy serves to increase survival time but is not curative
Accuracy of apparent diffusion coefficients and enhancement ratios on magnetic resonance imaging in differentiating primary cerebral lymphomas from glioblastoma
No full textBACKGROUND AND AIM: Gingival hyperpigmentation is an esthetic problem. The aim of the present study was to identify most effective treatment modality for managing generalized physiological gingival pigmentation. BACKGROUND AND PURPOSE: This study aimed to determine the accuracy of apparent diffusion coefficient (ADC) and enhancement ratio (ER) in discriminating primary cerebral lymphomas (PCL) and glioblastomas. MATERIALS AND METHODS: Circular regions of interest were randomly placed centrally within the largest solid-enhancing area of all lymphomas and glioblastomas on both post-contrast T1-weighted images and corresponding ADC maps. Regions of interest were also drawn in the contralateral hemisphere to obtain enhancement and ADC values of normal-appearing white matter. This helped us to calculate the ER and ADC ratio. RESULTS: Mean enhancement and ADC (mm2/s) values for PCL were 2220.56 ± 2948.30 and 712.00 ± 137.87, respectively. Mean enhancement and ADC values for glioblastoma were 1537.07 ± 1668.33 and 1037.93 ± 280.52, respectively. Differences in ADC values, ratios and ERs were all statistically significant between the two groups (p \u3c 0.05). ADC values correctly predicted 71.4% of the lesions as glioblastoma and 83.3% as PCL (area under the curve (AUC) = 0.86 on receiver operating characteristic curve analysis). ADC ratios correctly predicted 85.7% of the lesions as glioblastoma and 100% as PCL (AUC = 0.93). ERs correctly predicted 71.4% of the lesions as glioblastoma and 88.9% as PCL (AUC = 0.92). The combination of ADC ratio and ER correctly predicted 100% tumour type in both patient subgroups
