Manual Tuning Research Articles

Optimizing SVM for argan tree classification using Sentinel-2 data: A case study in the Sous-Massa Region, Morocco

The development of efficient classifiers for land cover remains challenging due to the presence of hyperparameters in the model. Conventional approaches rely on manual tuning, which is both time-consuming and impractical, often leading to suboptimal results. This study aimed to optimize the hyperparameters of the Support Vector Machine (SVM) algorithm using the grid search method to map the distribution of the Argan forest in the Souss-Massa region of Morocco from Sentinel-2 satellite image. To achieve this, we examined the C parameter for the linear function, as well as the C and gamma parameters for the radial RBF and sigmoid functions. Similarly, we explored the C, gamma, and degree parameters for the polynomial function chosen using the grid search method. These parameters are compared with the default hyperparameters of each SVM function. The results are validated using the cross-validation method and by the following scores: accuracy, precision, recall, F1 score, and Cohen’s Kappa. The experiments were conducted using the Earth Engine Python API in Google Colab (Google Collaboratory). In addition, experimental results indicate that the hyperparameters selected by grid search yield higher scores than the default hyperparameters. The best results were achieved using the hyperparameters of the polynomial base kernel, specifically with C = 10, degree = 2, and gamma = 10. Accuracy = 96.61%.

Deep Learning-Driven Analysis of a Six-Bar Mechanism for Personalized Gait Rehabilitation

Abstract Recent advances in robotics and artificial intelligence have highlighted the potential for the integration of computational intelligence in enhancing the functionality and adaptability of robotic systems, particularly in rehabilitation. Designing robotic exoskeletons for the lower limb rehabilitation of post-stroke patients requires frequent adjustments to accommodate individual differences in leg anatomy. This complex engineering challenge necessitates a deep understanding of human physiology, robotics, and optimization to develop adaptive robotic systems and also to swiftly quantify the required adjustments and implement them for each patient. The conventional approaches, which mostly rely on heuristics and manual tuning, often struggle to achieve optimal results. This paper presents a novel method that integrates a genetic algorithm (GA) with a deep learning approach to generate a gait trajectory of the ankle joint from a six-bar linkage mechanism of fixed dimensions. Later, using the same approach, the inverse kinematics solution for this mechanism is also devised whereby, the set of the link dimensions of the six-bar linkage mechanism is obtained for the given gait trajectory of an individual to achieve customization. We simulated the kinematic behavior of the six-bar linkage mechanism within defined mechanical constraints and utilized the generated data for training a feedforward neural network and long short-term memory models. The proposed model, when trained, can produce accurate lengths for the desired gait trajectories in the sagittal plane and vice-versa which further validates our proposed approach for inverse kinematics solution.

Recent Patents on Piano Tuning System

Piano tuning plays a very important role in piano playing and music education. Regular tuning can prolong the service life of the piano, improve the sound saturation, and make the sound more concentrated and full. Automated mechanical equipment, controllers, sensors, and other equipment are used to automatically tune pianos to solve the problems of human error, overtone resonance interference, string breakage, etc., occurring in manual tuning. To solve the above problems, based on the synthesis of related patents and research results, the patents on automatic tuning system and methods for modulating piano are introduced in detail. An automatic tuning system can replace manual tuning, simplify the complex process of tuning, and consider the problem of overtone interference caused by the vibration of the strings to improve the accuracy of tuning; it can, at the same time, perform real-time monitoring of the string tension, protecting the strings from breaking due to tuning. The automatic tuning system has a simple structure and a high degree of automation; its method is highly accurate and it can replace professional tuning technicians, being highly innovative and widely practical.

EI-ISOA-VMD: Adaptive denoising and detrending method for nuclear circulating water pump impeller

As a key component of the nuclear circulating water pump, it is difficult to extract effective information from low-quality impeller signal because of the interference noise and non-stationary trend, such as water flow impact, vibrations from unrelated components, and sensor noise. Traditional signal denoising methods, particularly those utilizing variational mode decomposition, often require manual parameter tuning, resulting in unsatisfactory outcome. To address these issues, this paper proposes a signal denoising and detrending method based on the evaluation index driven the improved seagull algorithm optimized for variational mode decomposition (EI-ISOA-VMD). During the optimization process, sinusoidal chaotic mapping is utilized for initializing the seagull population, while a tanh function is employed to design nonlinearly decreasing inertia weights. Furthermore, the Levy flight mechanism enhances the randomness of position updates and optimizes seagull individuals beyond the boundary range in order to form the improved seagull optimization algorithm. The denoising, detrending process, and evaluation index proposed in this paper are utilized as the fitness function for the improved seagull optimization algorithm to optimize key parameters of variational mode decomposition. During the denoising and detrending process, the trend component is initially selected using the mean ratio method. The useful, low noise and high noise intrinsic mode functions are screened by correlation coefficient and weighted permutation entropy. Subsequently, the low noise components are denoised by wavelet thresholding method, and the high noise components are discarded to finally reconstruct the signal. The experimental results demonstrate that the proposed method effectively eliminates noise and trend components in low-quality signal while preserving more valuable information.

Automated Neural Architecture Search for Cardiac Amyloidosis Classification from [18F]-Florbetaben PET Images.

Medical image classification using convolutional neural networks (CNNs) is promising but often requires extensive manual tuning for optimal model definition. Neural architecture search (NAS) automates this process, reducing human intervention significantly. This study applies NAS to [18F]-Florbetaben PET cardiac images for classifying cardiac amyloidosis (CA) sub-types (amyloid light chain (AL) and transthyretin amyloid (ATTR)) and controls. Following data preprocessing and augmentation, an evolutionary cell-based NAS approach with a fixed network macro-structure is employed, automatically deriving cells' micro-structure. The algorithm is executed five times, evaluating 100 mutating architectures per run on an augmented dataset of 4048 images (originally 597), totaling 5000 architectures evaluated. The best network (NAS-Net) achieves 76.95% overall accuracy. K-fold analysis yields mean ± SD percentages of sensitivity, specificity, and accuracy on the test dataset: AL subjects (98.7 ± 2.9, 99.3 ± 1.1, 99.7 ± 0.7), ATTR-CA subjects (93.3 ± 7.8, 78.0 ± 2.9, 70.9 ± 3.7), and controls (35.8 ± 14.6, 77.1 ± 2.0, 96.7 ± 4.4). NAS-derived network performance rivals manually determined networks in the literature while using fewer parameters, validating its automatic approach's efficacy.

Heat-Equation-Based Smoothing Homotopy Method for Nonlinear Optimal Control Problems

This paper presents a new heat-equation-based smoothing homotopy method for solving nonlinear optimal control problems with the indirect method. The surrogates, derived from the heat equation solution, are first incorporated into the necessary conditions as replacement of the terminal state and costate variables. The homotopy process is then applied to the heat conduction time: a longer time results in a more uniform temperature distribution and greater stability against initial temperature variations, thereby making the corresponding homotopy problem much easier to solve; zero time implies no heat conduction and reverts to the original problem. Furthermore, capitalizing on the heat equation’s boundedness characteristic, an efficient approach for determining the hypersensitive parameters is proposed, thus obviating the necessity of manual tuning. Challenging numerical examples are provided to demonstrate the superior performance of the proposed method, indicating that its convergence and efficiency are significantly enhanced compared to the original smoothing homotopy methods.

CONTROL SYSTEMS SYNTHESIS FOR ROBOTS ON THE BASE OF MACHINE LEARNING BY SYMBOLIC REGRESSION

This paper presents a novel numerical method for solving the control system synthesis problem through the application of machine learning techniques, with a particular focus on symbolic regression. Symbolic regression is used to automate the development of control systems by constructing mathematical expressions that describe control functions based on system data. Unlike traditional methods, which often require manual programming and tuning, this approach leverages machine learning to discover optimal control solutions. The paper introduces a general framework for machine learning in control system design, with an emphasis on the use of evolutionary algorithms to optimize the generated control functions. The key contribution of this research lies in the development of an algorithm based on the principle of small variations in the baseline solution. This approach significantly enhances the efficiency of discovering optimal control functions by systematically exploring the solution space with minimal adjustments. The method allows for the automatic generation of control laws, reducing the need for manual coding, which is especially beneficial in the context of complex control systems, such as robotics. To demonstrate the applicability of the method, the research applies symbolic regression to the control synthesis of a mobile robot. The results of this case study show that symbolic regression can effectively automate the process of generating control functions, significantly reducing development time while improving accuracy. However, the paper also acknowledges certain limitations, including the computational demands required for symbolic regression and the challenges associated with real-time implementation in highly dynamic environments. These issues represent important areas for future research, where further optimization and hybrid approaches may enhance the method's practicality and scalability in real-world applications.

Predictive modeling of evoked intracranial EEG response to medial temporal lobe stimulation in patients with epilepsy

Despite promising advancements, closed-loop neurostimulation for drug-resistant epilepsy (DRE) still relies on manual tuning and produces variable outcomes, while automated predictable algorithms remain an aspiration. As a fundamental step towards addressing this gap, here we study predictive dynamical models of human intracranial EEG (iEEG) response under parametrically rich neurostimulation. Using data from n = 13 DRE patients, we find that stimulation-triggered switched-linear models with ~300 ms of causal historical dependence best explain evoked iEEG dynamics. These models are highly consistent across different stimulation amplitudes and frequencies, allowing for learning a generalizable model from abundant STIM OFF and limited STIM ON data. Further, evoked iEEG in nearly all subjects exhibited a distance-dependent pattern, whereby stimulation directly impacts the actuation site and nearby regions (≲ 20 mm), affects medium-distance regions (20 ~ 100 mm) through network interactions, and hardly reaches more distal areas (≳ 100 mm). Peak network interaction occurs at 60 ~ 80 mm from the stimulation site. Due to their predictive accuracy and mechanistic interpretability, these models hold significant potential for model-based seizure forecasting and closed-loop neurostimulation design.

Prediction of Shale Gas Well Productivity Based on a Cuckoo-Optimized Neural Network

Current shale gas well production capacity predictions primarily rely on analytical and numerical simulation methods, which necessitate extensive calculations and manual parameter tuning and produce lowly accurate predictions. Although employing neural networks yields highly accurate predictions, they can easily fall into local optima. This paper suggests a new way to use Cuckoo Search (CS)-optimized neural networks to make shale gas well production capacity predictions more accurate and to solve the problem of local optima. It aims to assist engineers in devising more effective development plans and production strategies, optimizing resource allocation, and reducing risk. The method first analyzes the factors influencing the production capacity of shale gas wells in a block located in western China through correlation coefficients. It identifies the main factors affecting the gas test absolute open flow as organic carbon content, small-layer passage rate, fracture pressure, acid volume, pump-in fluid volume, brittle mineral content in the rock, and rock density. Subsequently, we used the CS algorithm to conduct the global training of the neural network, avoiding the problem of local optima, and established a neural network model for predicting shale gas well production capacity optimized by the CS algorithm. A comparative analysis with other relevant methods demonstrates that the CS-optimized neural network model can accurately predict production capacity, enabling a more rational and effective exploitation of shale gas resources, which lower development costs and increase the economic returns of oil and gas fields. Compared to numerical simulation, SVM, and BP neural network algorithms, the CS-optimized BP neural network (CS-BP) exhibits significantly lower prediction error. Its correlation coefficient between predicted and actual values reaches as high as 0.9924. Verification experiments conducted on another shale gas well also demonstrate that, in comparison to the BP neural network algorithm, CS-BP offers superior prediction performance, with model validation showing a prediction error of only 0.05. This study can facilitate more rational and efficient exploitation of shale gas resources, reduce development costs, and enhance the economic benefits of oil and gas fields.

Application of improved fish school algorithm in variable frequency speed control system

In order to solve the problems of long adjustment time and poor stability of the commonly used speed current variable frequency negative feedback speed control system in engineering, the author proposes an improved fish school algorithm application method in variable frequency speed control systems. A fish swarm algorithm optimized based on the arena method is applied to a variable frequency speed control system, and the optimization algorithm is applied to the speed control system to screen the PI parameters that meet the requirements of the speed control system. The simulation results show that compared with manual parameter tuning, the controller parameters optimized by the improved fish school algorithm have better control performance. In the absence of overshoot starting, the starting time was shortened by 0.21s, and the system response speed was improved. When a sudden load of 6 N m is applied, the speed drop is reduced by 3 r · min−1, and the recovery time is shortened by 0.15 s, resulting in stronger anti-interference performance of the system. Conclusion: The new algorithm shortens the control time and improves the robustness of the system.

Hyperparameter Optimization for Convolutional Neural Network-Based Sentiment Analysis

The rapid development of digital technology brings about the importance of product or service opinions as a data source in business intelligence. Sentiment analysis is a business intelligence tool that can help gauge market trends by analysing the available opinions. Various studies have been developed to solve sentiment analysis problems, from selecting feature engineering techniques to selecting sentiment classification models based on classical or deep learning methods. However, these studies still select hyperparameters on a trial basis, which poses problems in terms of time and not optimal performance. Therefore, this research proposes the implementation of Bayesian Optimization to automate the selection of hyperparameters in sentiment analysis based on a Convolutional Neural Network (CNN). The results show that implementing the Bayesian Optimization method has the lowest computation time compared to Random and Grid Searches, which is only 175.746765 seconds, in the fifth trial. In addition, it made a reasonably large cut in terms of times compared to manual trial-based hyperparameter tuning that needs 540 trials. In the model assessment process, implementing Bayesian Optimization gives the highest values on Precision and Specificity, which are 0.8168 and 0.7176, respectively. Thus, implementing Bayesian Optimization is the right choice for sentiment analysis tasks since the precision of predicting negative sentiment classes is vital in hospitality business intelligence, especially the use of related information for product improvement or hotel service improvement. The implementation of Bayesian Optimization not only applied as a reliable hyperparameter selection technique for classification tasks but also regression prediction tasks.

PPG2RespNet: a deep learning model for respirational signal synthesis and monitoring from photoplethysmography (PPG) signal.

Breathing conditions affect a wide range of people, including those with respiratory issues like asthma and sleep apnea. Smartwatches with photoplethysmogram (PPG) sensors can monitor breathing. However, current methods have limitations due to manual parameter tuning and pre-defined features. To address this challenge, we propose the PPG2RespNet deep-learning framework. It draws inspiration from the UNet and UNet + + models. It uses three publicly available PPG datasets (VORTAL, BIDMC, Capnobase) to autonomously and efficiently extract respiratory signals. The datasets contain PPG data from different groups, such as intensive care unit patients, pediatric patients, and healthy subjects. Unlike conventional U-Net architectures, PPG2RespNet introduces layered skip connections, establishing hierarchical and dense connections for robust signal extraction. The bottleneck layer of the model is also modified to enhance the extraction of latent features. To evaluate PPG2RespNet's performance, we assessed its ability to reconstruct respiratory signals and estimate respiration rates. The model outperformed other models in signal-to-signal synthesis, achieving exceptional Pearson correlation coefficients (PCCs) with ground truth respiratory signals: 0.94 for BIDMC, 0.95 for VORTAL, and 0.96 for Capnobase. With mean absolute errors (MAE) of 0.69, 0.58, and 0.11 for the respective datasets, the model exhibited remarkable precision in estimating respiration rates. We used regression and Bland-Altman plots to analyze the predictions of the model in comparison to the ground truth. PPG2RespNet can thus obtain high-quality respiratory signals non-invasively, making it a valuable tool for calculating respiration rates.

Risk identification and prediction model for continuous‐lane‐change vehicles considering driving style

Lane change risk identification and prediction research is important for traffic safety, development of advanced driver assistance systems, traffic flow optimization. Among them, vehicle continuous‐lane‐change (CLC) risk identification and prediction is weak, but this element is important for vehicle decision-making and path planning in multi-lane, non-adjacent lane situations. To compensate for the shortcomings of risk identification based on both deterministic and probabilistic methods and address the issues of poor measurement accuracy and efficiency caused by manual tuning, this study established a risk identification and prediction model based on the Bayesian global optimization (BO) gated recurrent unit (GRU) neural network (BO-GRU) and instantiated the model using the trajectory data of 129 vehicles. Overall, the results showed that the accuracy of identifying CLC risks and driving-style was 99.33% and 97.44%, respectively, while the overall average accuracy of predicting potential CLC risks based on the identification of risks and predicted trajectories was 97.13%. The root-mean-square errors of CLC predicted trajectory for different driving-styles were 0.2503, 0.2331, 0.2093, and 0.2706, respectively, which were 84%−88% lower compared to 1.7361 when driving-styles were not considered. The performance of the BO-GRU model is comparable to that of parameter-optimized LSTM and its variant models, which highlights the significance of parameter optimization for neural network models. Overall, the use of a BO-GRU-based model for identifying and predicting continuous lane change risks in vehicles can aid in making driving decisions and planning routes, decrease the risk of collisions, enhance traffic flow efficiency, and encourage safe driving behavior.

Optimization of machine learning model training procedure on multi-gpu systems to enhance cyber security in telecommunication networks

This paper analyzes the optimization features of machine learning (ML) model training procedures using multi-GPU systems to enhance cyber security in telecommunication networks. A key aspect of the study is the use of data parallelism, which allows the distribution of the training load across multiple GPUs, significantly reducing training time and improving model accuracy-critical factors for rapid threat detection in cyberspace. A novel approach for optimizing data batch size using Mutual Information (MI) is proposed, which harmonizes the utilization of computational resources with the information content of the training data. MI helps to determine the optimal data batch size that minimizes training errors and improves model accuracy without a significant increase in training time. Experimental results demonstrate the substantial advantages of multi-GPU configurations compared to single-GPU setups, providing faster training and improved model accuracy. It was particularly emphasized that MI-guided batch size tuning significantly outperforms traditional manual tuning methods, ensuring higher validation accuracy and reducing training time. The study showed that the MI-based approach is an effective tool for addressing the problem of optimizing ML model training processes in real-world scenarios where cyber security is critical. The proposed methods allow ML models to train faster and more accurately identify potential threats, making them particularly relevant for telecommunication networks where a rapid response to new threats in real time is required. The implementation of modern computational technologies such as multi-GPU systems and MI-optimized training enhances the efficiency and accuracy of machine learning models. This, in turn, improves cyber security measures and ensures a more reliable defence of telecommunication networks against malicious attacks. It is noted that the proposed approaches can be adapted not only for cyber security but also for other areas where high model accuracy and fast training are important. Future research prospects include the development of new machine learning methods, particularly deep neural networks, the exploration of alternative computational architectures such as quantum computing or distributed systems, and their integration into real-time systems. Special attention should be paid to the ethical aspects of implementing automated cyber security systems, particularly in preventing bias in algorithms and ensuring fairness in their application.

PI gain tuning for pressure-based MFCs with Gaussian mixture model

A vast number of mass flow controllers (MFCs) are used in semiconductor industry. For the stable supply, an efficient production method of MFC is required. The gain tuning of the proportional-integral (PI) control to realize a setting flow rate is essential for efficient mass production. The gains are tuned to meet the specifications required for evaluation indices of response time and overshoot amount in a step response waveform. The tuning is complicated especially for the case of pressure-based MFCs. In this paper, we propose a simple method for the PI gain tuning using the Gaussian mixture model (GMM) and the direct inverse analysis applicable to the pressure-based MFCs’ production. The relationship between the gains and evaluation indices for a standard unit of the MFC is modeled as the GMM. The direct inverse analysis calculates the difference between the standard and a test unit. Under the assumption that the difference can be compensated by a simple shift, gains likely to meet the specifications for the test unit are searched. We applied the method to seven test units. The result showed that the gains of all the test units were tuned within only a few iterations whose numbers were much less than the conventional manual tuning method, and there was no untunable unit.

Clustering-based hyper-heuristic algorithm for multi-region coverage path planning of heterogeneous UAVs

In the context of multi-heterogeneous UAV coverage path planning, an effective solution method has been proposed. Firstly, regions are set up as fully connected graphs which are cut into multiple subgraphs by spectral clustering method to assign tasks to multi-heterogeneous UAVs. Additionally, an RL-based hyper-heuristic algorithm is proposed. Heuristic space is parameterized by GNN which is trained with the reward provided by the optimization goal to automate design and enhance the heuristic metrics, avoiding the inefficiency and suboptimality of expert design and manual parameter tuning. Compared with existing methods, the proposed algorithm has a better performance in task completion time, execution time and deviation rate, which shows its potential application in the coverage path planning problem of multi-heterogeneous UAVs.

Sigmoidal learning rate optimizer for deep neural network training using a two-phase adaptation approach

The optimization of machine learning models is an open research question since an optimal procedure to change the learning rate throughout training is still unknown. Manually defining a learning rate schedule involves troublesome, time-consuming try-and-error procedures to determine hyperparameters, such as learning rate decay epochs and learning rate decay rates, in order to navigate the intricate landscape of loss functions. Although adaptive learning rate optimizers automatize this process, recent studies suggest they may produce overfitting and reduce performance compared to fine-tuned learning rate schedules. Considering that the new machine learning loss function approaches present landscapes with much more saddle points than local minima, we proposed the Training Aware Sigmoidal Optimizer (TASO). This automated two-phase learning rate adaptation mechanism significantly reduces the need for manual hyperparameter tuning. The first phase uses a high learning rate to quickly traverse the numerous saddle points in the error surface, while the second phase uses a low learning rate to gradually approach the center of the local minimum previously found. We compared the proposed approach with commonly used adaptive learning rate schedules such as Adam, RMSProp, and Adagrad. The validation experiments were performed in image and text datasets and showed that TASO outperformed all competing methods in optimal (i.e., performing hyperparameter validation) and suboptimal (i.e., using default hyperparameters) scenarios. In our benchmark tests, TASO demonstrated promising performance, achieving an 8.32% increase in accuracy and a significant 46.62% decrease in training loss on average across various datasets and models. This remarkable performance positions TASO ahead of well-established adaptive optimizers, suggesting higher effectiveness and consistency in performance.

Detecting the interaction between microparticles and biomass in biological wastewater treatment process with Deep Learning method

Investigating the interaction between influent particles and biomass is basic and important for the biological wastewater treatment. The micro-level methods allow for this, such as the microscope image analysis method with the conventional ImageJ processing software. However, these methods are cost and time-consuming, and require a large amount of work on manual parameter tuning. To deal with this problem, we proposed a deep learning (DL) method to automatically detect and quantify microparticles free from biomass and entrapped in biomass from microscope images. Firstly, we introduced a “TU Delft-Interaction between Particles and Biomass” dataset containing labeled microscope images. Then, we built DL models using this dataset with seven state-of-the-art model architectures for a instance segmentation task, such as Mask R-CNN, Cascade Mask R-CNN, Yolact and YOLOv8. The results show that the Cascade Mask R-CNN with ResNet50 backbone achieves promising detection accuracy, with a mAP50box and mAP50mask of 90.6 % on the test set. Then, we benchmarked our results against the conventional ImageJ processing method. The results show that the DL method significantly outperforms the ImageJ processing method in terms of detection accuracy and processing cost. The DL method shows a 13.8 % improvement in micro-average precision, and a 21.7 % improvement in micro-average recall, compared to the ImageJ method. Moreover, the DL method can process 70 images within 1 min, while the ImageJ method costs at least 6 h. The promising performance of our method allows it to offer a potential alternative to examine the interaction between microparticles and biomass in biological wastewater treatment process in an affordable manner. This approach offers more useful insights into the treatment process, enabling further reveal the microparticles transfer in biological treatment systems.

Heat spreading effect on the optimal geometries of cooling structures in a manifold heat sink

Embedded cooling is a promising solution to address thermal challenges of power electronics. The present work investigates heat spreading effect on the optimal geometries of cooling structures within a single-phase manifold heat sink in laminar regime by integration of topology optimization (TO) and Bayesian optimization (BO). Such integration is for the manual parameter tuning issue in the density-based TO of conjugate heat transfer, and the TO process is accelerated by using a dual-level simulator and the optimality criteria optimizer combined with the stochastic gradient descent method. The heat spreading effect significantly alters the optimal geometries, due to the change of thermal resistance constitutions. Physical analyses on the BO-TO-generated geometries give two guidelines for improving the cooling structures here: (a) Heat source-coolant distance should increase properly for narrow heating area and small Biot number; (b) Highly-thermal-conductivity solid can be filled into low-velocity regions directly connecting to hot area. As an example, a fabrication-friendly cooling structure is devised in the case of concentrating heating following these guidelines, avoiding the computationally-expensive optimization process and the over-complicated geometries. The CFD simulations show this design can significantly reduce the thermal resistance with minor increase of pressure loss as the Reynold number from 50 to 300.

Optimal channel assignment on dense Wi-Fi networks using Thermodynamic Threshold Accepting

Channel assignment in Wi-Fi 6 dense networks is a challenging problem which critically impacts the performance of WLAN communications. Although there have been many proposals in the area of IEEE 802.11 channel assignment, LCCS has emerged as the de facto standard for distributed channel selection. Notwithstanding, centralized optimizers overcome the performance of distributed channel assignment algorithms, so their study is of utmost importance. One of the most successful optimization proposals in the state-of-the-art for Wi-Fi 4 is based on simulated annealing (SA). However, SA presents a significant limitation: its parameters must be adjusted manually for each scenario, otherwise yielding suboptimal solutions. In this paper, we propose a new technique for channel assignment optimization of Wi-Fi dense networks, called Thermodynamic Threshold Accepting (TTA), including modifications inspired by Threshold Accepting (TA) and Thermodynamic Simulated Annealing (TSA). With respect to SA, TTA avoids manual tuning, being able to adapt to the specific setting automatically. We evaluate TTA in a dense Wi-Fi model and, compared to SA, we show that our approach reaches optimal results without requiring any prior fine-tuning. We also show that TTA is even able to slightly overcome the performance of an accurately calibrated SA optimizer for the same number of iterations.