• Collected a state-of-the-art dataset, focussing on activities of daily living (ADL) primarily centered around finger movements, under different measurement conditions.
• Employed extensive evaluations to gauge the efficacy of ML models across a wide spectrum of experimental scenarios and setups.
• FCNN and EMGHandNet as the top-performing models with remarkable test accuracies of 88% and 78.63%, respectively.
• EMGHandNet exhibited a notable performance improvement over FCNN, achieving a remarkable 27.47% higher accuracy for varying body postures.

• Conducted thorough analysis of established Nanomagnetic Logic (NML) simulators, like ToPoliNano, MagCad, and NMLSim2.0.
• Designed and simulated simple circuits (Majority Gates, Full Adders) using these simulators for comprehensive analysis.
• Recognized NMLSim2.0 as the prominent simulator, owing to its circuit simulation capabilities, and inclusion of cuts.

Vidya Sagar, Anish Turlapaty, Surya Naidu

• The new dataset, EMAHA-DB4 includes sEMG signals from 10 subjects performing 8 unique ADL in 3 body postures and 4 arm positions for varied measurement conditions.
• The framework uses features from time domain, frequency domain, wavelet domain, and Eigenvalues for ML classification.
• CNN Bi-LSTM achieved 85.37% accuracy for activities versus arm positions and 82.1% versus body postures in aggregate scenarios.
• DNN consistently achieved higher accuracy with averages of 90.9% in aggregate conditions and 90.3% in subject-specific scenarios.

Eswar Adapa, Anish Turlapaty, Surya Naidu

• Addresses muscle fatigue detection and onset estimation during strength training using sEMG signals.
• Diverse measurement conditions (postures, experience levels, lifting loads) are considered for method evaluation.
• EMAHAD-DB6 dataset created, featuring sEMG signals from 11 subjects in various strength training scenarios.
• SVM classifier outperforms others with an 86.5% test accuracy, especially excelling in low load conditions (94% accuracy). Fatigue onset estimation has a 12% average relative error.

Sayee Sreenivas, Anish Turlapaty, Surya Naidu

• First of its kind dataset focussing on rate of doing activites using sEMG signals.
• Employed Signal Processing techniques to extract relevant features from the signals.
• Utilizing ML techniques to classify the activites under different experimental setups.

Anish Turlapaty, Surya Naidu, Vidya Sagar

• Class-wise Analysis of EMAHA-DB5 is perfromed to understand the impact of various classes using F1- score.
• % of Train Data and the impact on Accuracy of model is also evaluated.
• Impact of training on various combinations of Body and Hand Postures is also analysed, giving very promising results.

Anish Turlapaty, Vidya Sagar, Surya Naidu

• Application of Transfer Learning on EMAHA-DB4 dataset.
• In-depth analysis of impact of measurement conditions on Model performance.
• Implementing U-Nets and other DL architectures for fine-tuning the model.

Eswar Adapa, Anish Turlapaty, Surya Naidu

• Fine-tuning the model to reduce average error.
• Data balancing to take a step further for real-time implementation.
• Deep Learning techniques like U-Nets are currently being examined.