Learning task-specific models for dexterous, in-hand manipulation with simple, adaptive robot hands

In dexterous robotic hand design, achieving high mobility and adaptability comparable to human hands remains an ongoing challenge. Biomimetic designs mimicking the musculoskeletal structure have shown promise yet face difficulties in preserving key kinematic and mechanical principles while reducing system complexity. Here, we present a biomimetic finger design that preserves these principles through coordinated rigid-soft interplay, achieving structural and control simplicity for constructing dexterous robotic hands. Our design distills complex anatomical structures into skeletal mechanisms with regular geometrics, strategically deployed soft ligaments, and elastic tendon actuation, enabling controllable multi-degree-of-freedom dexterity while providing resilience and compliance. We establish mathematical models to analyze finger kinematics, rigid-soft interplay principles, and controllable actuation. Building on these models, we integrate biomimetic fingers with a thumb to develop an anthropomorphic robotic hand. Our robotic hand experimentally demonstrates remarkable dexterity and versatility across various tasks, including piano playing, power and pinch grasping, and in-hand manipulation, confirming the design effectiveness.

Machine learning (ML) approaches to predicting quantum mechanical (QM) properties have made great strides toward achieving the computational chemist's holy grail of structure-based property prediction. In contrast to direct ML methods, which encode a molecule with only structural information, in this work, we show that QM descriptors improve ML predictions of dimer interaction energy, both in terms of accuracy and data efficiency, by incorporating electronic information into the descriptor. We present the electron deformation density interaction energy machine learning (EDDIE-ML) model, which predicts the interaction energy as a function of Hartree-Fock electron deformation density. We compare its performance with leading direct ML schemes and modern DFT methods for the prediction of interaction energies for dimers of varying charge type, size, and intermolecular separation. Under a low-data regime, EDDIE-ML outperforms other direct ML schemes and is the only model readily transferrable to larger, more complex systems including base pair trimers and porous cages. The underlying physical connection between the density and interaction energy enables EDDIE-ML to reach an accuracy comparable to modern DFT functionals in fewer training data points compared to other ML methods.

Brain tumor implies development of strange cells in brain. In cutting edge stages, brain tumor is most risky infection which can't be relieved. Thus, it ought to be identified in the beginning phases with the assistance of MRI (Magnetic Resonance Image). So, there will be more changes to the patient to endure. Quite possibly, the most functional and significant technique is to utilize Deep Neural Network (DNN). In this paper, a Deep Convolutional Neural Network (DCNN) based on VGG16 model has been developed to identify a tumor through brain Magnetic Resonance Imaging (MRI) dataset. In the clinical field, the strategies of machine learning (ML) and data mining hold a critical stand. It is effectively used to achieve the efficiency and exact location of tumor. The proposed technique involves automatic segmentation method based on Deep Convolution neural network (DCNN). It is layer based segmentation and classification technique. Different levels are engaged with the proposed technique, first step is data collection and then pre-processing & average filtering is done, after that segmentation, feature extraction and classification via DCNN is performed. In this paper, a fresh technique for classification of brain MRI images has been proposed using DCNN based softmax Classifier. The proposed system has been compared with the existing ones. The results are really encouraging as the loss is reduced and accuracy is increased. The proposed system achieved the training accuracy of 98.9% and validation accuracy of 100%. The training loss is reduced up to 0.0230 and the validation loss reduced up to 0.0109.

Chapter 16 - Prediction of leukemia by classification and clustering techniques

Pathological high frequency oscillations (HFOs) are putative neurophysiological biomarkers of epileptogenic brain tissue. Utilizing HFOs for epilepsy surgery planning offers the promise of improved seizure outcomes for patients with medically refractory epilepsy. This review discusses possible machine learning strategies that can be applied to HFO biomarkers to better identify epileptogenic regions. We discuss the role of HFO rate, and utilizing features such as explicit HFO properties (spectral content, duration, and power) and phase-amplitude coupling for distinguishing pathological HFO (pHFO) events from physiological HFO events. In addition, the review highlights the importance of neuroanatomical localization in machine learning strategies.

This research work focuses on detecting cyberthreats, including Malware, Man-in-the-Middle, and Distributed Denial-of-Service. Given the escalating complexity and frequency of cyber threats, traditional detection approaches struggle to provide adequate protection, underscoring the need for deep learning and machine learning strategies. This study evaluates and compares the performance of six prominent models: XGBoost, Random Forest, Support Vector Machine, Dense Neural Network, Convolutional Neural Network, and Long Short-Term Memory using the NF-UQ-NIDS-v2 Network Intrusion Detection Dataset. The dataset was meticulously preprocessed using the scaling technique and feature selection with the Chi-square method (selecting 20 features out of 43), as well as Random Oversampling to address class imbalance. Experimental evaluations utilize a group of classification metrics to evaluate the effectiveness of models. The received scores demonstrated that ensemble models, specifically Random Forest and XGBoost, significantly outperformed other classifiers, achieving nearly perfect metrics of 0.999, which reflects their exceptional capability to generalize across diverse attack types. In contrast, the Support Vector Machine demonstrated limited effectiveness, with an accuracy as low as 0.794, indicating its relative inefficiency in this scenario. Deep learning models have consistently shown strong performance with metrics ranging from 0.924 to 0.949. This study highlights the effectiveness of ensemble techniques in cybersecurity threat classification, offering valuable insights for enhancing proactive defense mechanisms in network security systems

The emergence of large language models (LLMs) has provided robust support for application tasks across various domains, such as name entity recognition (NER) in the general domain. However, due to the particularity of the medical domain, the research on understanding and improving the effectiveness of LLMs on biomedical named entity recognition (BNER) tasks remains relatively limited, especially in the context of Chinese text. In this study, we extensively evaluate several typical LLMs, including ChatGLM2-6B, GLM-130B, GPT-3.5, and GPT-4, on the Chinese BNER task by leveraging a real-world Chinese electronic medical record (EMR) dataset and a public dataset. The experimental results demonstrate the promising yet limited performance of LLMs with zero-shot and few-shot prompt designs for Chinese BNER tasks. More importantly, instruction fine-tuning significantly enhances the performance of LLMs. The fine-tuned offline ChatGLM2-6B surpassed the performance of the task-specific model BiLSTM+CRF (BC) on the real-world dataset. The best fine-tuned model, GPT-3.5, outperforms all other LLMs on the publicly available CCKS2017 dataset, even surpassing half of the baselines; however, it still remains challenging for it to surpass the state-of-the-art task-specific models, i.e., Dictionary-guided Attention Network (DGAN). To our knowledge, this study is the first attempt to evaluate the performance of LLMs on Chinese BNER tasks, which emphasizes the prospective and transformative implications of utilizing LLMs on Chinese BNER tasks. Furthermore, we summarize our findings into a set of actionable guidelines for future researchers on how to effectively leverage LLMs to become experts in specific tasks.

In the application of quantum communication networks, it is an important task to realize the optimal allocation of resources according to the current situation. For example, We need to select the optimal quantum key distribution (QKD) protocol and parameters. Traditionally, the most commonly implemented method is the local search algorithm (LSA), which costs a lot of resources. Here in this work, we propose a machine learning based scheme, in which the regression machine learning is used to simultaneously select the optimal protocol and corresponding parameters. In addition, we make comparisons among a few machine learning models including random forest (RF), K-nearest neighbor (KNN) and logistic regression. Simulation results show that the new scheme takes much less time than the LSA scheme, and the RF achieves the best performance. In addition, through the RF residual analysis, we find that the machine learning scheme has good robustness. In conclusion, this work may play an important role in promoting the practical application of quantum communication networks.

A flood is a most destructive disaster that affects people, places, and lives. Due to the complication in data availability, flood prediction is always a challenging task. The conventional mode of disaster management relies on satellite images and radar outcomes. It takes enormous time for processing. Machine learning paved the way for a new perspective on this hydrological problem. Recent developments in Machine Learning (ML) and Information and Communication Technology (ICT) have led to a state-of-the-art implementation and prediction. The major objective of this work is to recognize the most accurate machine learning model to identify flood occurrence, by comparing Logistic regression, Decision Tree, Naive Bayes, and Support Vector Machines classifiers. Machine Learning strategies are evaluated using precision, recall, F1-score, RMSE, and accuracy metrics. All the strategies are applied to one-feature dataset, three-feature dataset and four-feature dataset. The quantitative evaluation demonstrates that decision tree algorithm is most suitable for flood prediction and it exponentially grows with respect to the number of features examined.

Machine learning and pattern recognition recently become a hot topic in computing world. This is due to the fast-growing of resources as well as techniques that make it easier to solve machine learning and pattern recognition problems. Problems that require machine learning solutions may be very simple for humans but actually can be very complex for machines to solve them. Face recognition is amongst those problems. Almost all human can easily recognize others without require specific knowledge to do it, different from machines which require its. This paper discussed face recognition task using machine learning strategies which involved Kernel Principal Component Analysis (KPCA) and Support Vector Machine (SVM) to identify person. KPCA extracted features from 2D image input and produced the important features of an image input. The extracted face features are recognized by SVM by classifying human face according to their stored identity in a database. SVM, which was basically a binary classifier worked by using one-against-one strategy to compare the face feature vector of a single test image to the stored face image in a face image database. Experiment results on grayscale images with size 92x112 pixels gave 96.25% of accuracy rate. Hence, KPCA and SVM for face recognition is a robust machine learning method.

The oil and gas business are shed light on by means of this paper's comprehensive state of the art analysis of machine learning and fake insights. Areas include condition-based support for improvement of form in refining, quality management downstream and AI uses in the enhancement of MS refineries. It then proceeds to explain different strategies of machine learning and manufactured insights that could be applied to other upstream areas of oil and gas industry for data preparation and conversion. Gains and improvements confirm that machine learning and fake insights strategies for progression bear fruit resulting in increased precision in calculations and doubled capacities and storage spaces for more extensive amounts of data. This article also presents a broad spectrum of the multiple analysts that have enabled or enhanced the upstream and downstream oil and gas sectors, including machine learning and synthetic insights, and their drawbacks.

Improved flood depth estimation with SAR image, digital elevation model, and machine learning schemes

[EMBARGOED UNTIL 12/1/2024] This dissertation reports three case studies using machine learning, biophysical, and automation frameworks to study neural engineering challenges. The first study utilized machine learning with a clinical dataset to predict the risk of future opioid use disorder (OUD). The model achieved a high level of predictive accuracy and highlighted the most impactful variables that predicted the risk. The second study implemented recent tract- tracing data predicting the existence of a motif that generated the theta rhythm, similar to that in the hippocampus in the amygdala. This was done via the development of a biophysical model of the rodent amygdala that demonstrated how the theta rhythm could be engendered by an external theta-rhythmic inhibitory projection from the ventral pallidum and substantia innominata. The third study developed an automation pipeline using biophysical and machine learning schemes, to assist in the development of biophysical models of neurons. The approach implemented recent insights developed in our group related to currents being grouped into modules based on their neurocomputational signatures.

Machine learning is a technique for training computers to be used in classification or forecasting. Among the various types, support vector machine (SVM) is a fast and reliable machine learning mechanism. In this paper, we evaluate the stock price predictability of SVM based on financial statements, through a fundamental analysis predicting the stock price from the corporate intrinsic values. Corporate financial statements were used as the input for SVM. Based on the results, the rise or drop of the stock was predicted. The SVM results were compared with the forecasts of experts, as well as other machine learning methods such as ANN, decision tree and AdaBoost. SVM showed good predictive power while requiring less execution time than the other machine learning schemes.

We introduce a translational and rotational invariant local representation for vector fields, which can be employed in the construction of machine learning energy models of solids and molecules. This allows us to describe, on the same footing, the energy fluctuations due to the atomic motion, the longitudinal and transverse excitations of the vector field, and their mutual interplay. The formalism can then be applied to physical systems where the total energy is determined by a vector density, as in the case of magnetism. Our representation is constructed over the power spectrum of the combined angular momentum describing the local atomic positions and the vector field, and it can be used in conjunction with different machine learning schemes and data taken from accurate ab initio electronic structure theories. We demonstrate the descriptive power of our representation for a range of classical spin Hamiltonian and machine learning algorithms. In particular, we construct energy models based on both linear Ridge regression, as in conventional spectral neighbor analysis potentials, and the Gaussian approximation. These are both built to represent a Heisenberg-type Hamiltonian including a longitudinal energy term and spin-lattice coupling.