Manual Designs Research Articles

Designing an adaptive learning framework for predicting drug-target affinity using reinforcement learning and graph neural networks

In the field of biomedical engineering, predicting Drug-Target Affinities (DTA) is crucial. However, current affinity prediction models are predominantly manually designed, which is a complex, time-consuming process that may not effectively accommodate the diversity and complexity of datasets. To address this challenge, we propose an adaptive learning framework for predicting drug-target affinity, called Adaptive-DTA, which integrates reinforcement learning with graph neural networks to automate the design of affinity prediction models. Adaptive-DTA defines the architecture search space using directed acyclic graphs and employs an reinforcement learning algorithm to guide the architecture search, optimizing parameters based on the entropy of sampled architectures and model performance metrics. Additionally, we enhance efficiency with a two-stage training and validation strategy, incorporating low-fidelity and high-fidelity evaluations. Our framework not only alleviates the challenges associated with manual model design but also significantly improves model performance and generalization. To evaluate the performance of our method, we conducted extensive experiments on DTA benchmark datasets and compared the results with nine state-of-the-art methods. The experimental outcomes demonstrate that our proposed framework outperforms these methods, exhibiting outstanding performance in predicting drug-target affinities. Our innovative approach streamlines the design of affinity prediction model, reduces reliance on manual crafting, and enhances model generalization. Its ability to automatically optimize network architectures represents a major step forward in the automation of computational drug discovery processes.

Design of 1.1T/300mm Bitter-type magnet for gyro-traveling-wave tube with power consumption optimization theory.

To improve the portability of magnets in gyrotron devices, we designed a compact Bitter-type magnet with power consumption optimization theory. This magnet operates at room temperature in a small volume. The theory revises existing electromagnetic theory for non-uniform structural Bitter-type magnets and achieves the lowest energy consumption through iterative optimization. To extend the magnetic field homogeneity region, the ferromagnetic material armature is applied to the Bitter-type system without additional power consumption. Unlike previous manual designs, the proposed Bitter-type magnets can obtain optimal parameters with a significant reduction in computing time. Through the introduction of correction factors, we improve accuracy through multiple verifications of simulations and experiments. On this basis, a room-temperature Bitter-type magnet system for Ka-band fundamental mode gyrotron amplifiers is designed. Its maximum magnetic field strength is 1.1T, and the length of the homogeneity region is 300mm. Through optimization, its energy consumption is only 27.5 kW.

FDE-Net: A memory-efficiency densely connected network inspired from fractional-order differential equations for single image super-resolution

Dense connection has proved an effective way to take full advantage of hierarchical features extracted from low-resolution images using deep neural networks (DNNs) for single image super-resolution (SISR). However, existing densely connected networks are often manually designed and overly depended on practical experience, thus leading to suboptimal performance. Moreover, due to their complicated connections, they are memory-consuming and lack of interpretability. To address all these problems with one stone, we propose to construct a new densely connected network for SISR from a dynamic system perspective. Following this idea, we cast the hierarchical transformation of DNNs into the state evolution of a fractional-order dynamic system, which empowers us to automatically construct two interdependent densely connected modules based on the system solution rather manual design. They are a prediction module that controls the system to iteratively predict the next state, and a correction module that iteratively refines the predicted state to improve the prediction accuracy. With these two modules as backbone, we establish a Fractional-order Differential Equations-based network (FDE-Net) for SISR. Since the iterative computational manner requires both densely connected modules to be of the recurrent structure, FDE-Net is memory-efficiency and good interpretability. In addition, we analyze the existence and uniqueness of the solution for FDE to theoretically guarantee the feasibility of FDE-Net. Experiments on four SISR benchmark datasets demonstrate the superiority of FDE-Net over existing densely connected networks and other baselines in terms of generalization capacity, especially with limited memory.

Assessing embodied carbon of flat slab buildings – An ANN-integrated optimization methodology

Currently, there is a scarcity of studies that comprehensively address all structural components in the building's overall environmental performance. Although mathematical optimization has shown its great effectiveness, the prevailing iterative design approach continues hindering the simultaneous optimization of building structural performance and its sustainability. This research aims to address this gap by introducing an innovative optimization framework that integrates optimization concepts into the design process using Artificial Neural Networks (ANNs). The proposed framework includes a carbon-optimization problem for flat slab buildings, which is validated with manual designs, and a surrogate modelling stage employing ANNs to predict optimal design solutions. The best network demonstrates remarkable predictive power, yielding minimal errors and high fitness (>0.96) for all variables, except slab thickness. Sensitivity analysis identifies the building's height and dimensions as the most influential factors on the slab reinforcement, column reinforcement, and carbon footprint. While span length and slab concrete strength greatly impact slab and drop depths, column size is primarily controlled by span length and building's height. This research not only provides crucial insights into sustainable design solutions but also paves a clear pathway towards achieving decarbonization and cleaner production across the entire building sector.

Hybrid-Supervised Dual-Search: Leveraging Automatic Learning for Loss-Free Multi-Exposure Image Fusion

Multi-exposure image fusion (MEF) has emerged as a prominent solution to address the limitations of digital imaging in representing varied exposure levels. Despite its advancements, the field grapples with challenges, notably the reliance on manual designs for network structures and loss functions, and the constraints of utilizing simulated reference images as ground truths. Consequently, current methodologies often suffer from color distortions and exposure artifacts, further complicating the quest for authentic image representation. In addressing these challenges, this paper presents a Hybrid-Supervised Dual-Search approach for MEF, dubbed HSDS-MEF, which introduces a bi-level optimization search scheme for automatic design of both network structures and loss functions. More specifically, we harness a unique dual research mechanism rooted in a novel weighted structure refinement architecture search. Besides, a hybrid supervised contrast constraint seamlessly guides and integrates with searching process, facilitating a more adaptive and comprehensive search for optimal loss functions. We realize the state-of-the-art performance in comparison to various competitive schemes, yielding a 10.61% and 4.38% improvement in Visual Information Fidelity (VIF) for general and no-reference scenarios, respectively, while providing results with high contrast, rich details and colors. The code is available at https://github.com/RollingPlain/HSDS_MEF.

Few-Shot Font Generation With Weakly Supervised Localized Representations.

Automatic few-shot font generation aims to solve a well-defined, real-world problem because manual font designs are expensive and sensitive to the expertise of designers. Existing methods learn to disentangle style and content elements by developing a universal style representation for each font style. However, this approach limits the model in representing diverse local styles because it is unsuitable for complicated letter systems. For example, Chinese characters consist of a varying number of components (often called "radical") with a highly complex structure. In this paper, we propose a novel font generation method that learns localized styles, namely component-wise style representations, instead of universal styles. The proposed style representations enable synthesizing complex local details in text designs. However, learning component-wise styles solely from a few reference glyphs is infeasible when a target script has a large number of components, for example, over 200 for Chinese. To reduce the number of required reference glyphs, we represent component-wise styles by a product of component and style factors inspired by low-rank matrix factorization. Owing to the combination of strong representation and a compact factorization strategy, our method shows remarkably better few-shot font generation results (with only eight reference glyphs) than other state-of-the-art methods. Moreover, strong locality supervision was not utilized, such as the location of each component, skeleton, or strokes. The source code is available at https://github.com/clovaai/lffont.

Evolving many-objective dispatching rule pairs for unrelated parallel machine scheduling with sequence-dependent setup times

The real-world problem of unrelated parallel machine scheduling problem with sequence-dependent setup times (UPMSPSST) is investigated and focuses on two key aspects: fast solution and many-objective optimization. While dispatching rules (DRs) are effective in meeting demand, the manual design of many-objective DRs (MaODRs) is challenging. An alternative approach, based on the automatic evolution of MaODRs using hyper-heuristics, shows promise. However, limited literature exists on the automatic evolution of MaODRs specifically for UPMSPSST. In this study, a many-objective genetic programming (MaOGP) algorithm is formed by combining many-objective evolutionary algorithms (MaOEAs) with genetic programming (GP) algorithms. This enables the automatic evolution of MaODRs, leveraging the respective advantages of each approach. The exploration results demonstrate that the evolved MaODRs outperform manually designed DRs reported in prior literature. Furthermore, when considering combinations of objectives with different correlations, the MaODRs exhibit outstanding performance across multiple objectives simultaneously, surpassing even the performance of DRs trained separately for individual objectives. These findings highlight the potential of the proposed MaOGP approach in efficiently and effectively solving the UPMSPSST problem, opening avenues for further research and practical applications in related industries.

LRNAS: Differentiable Searching for Adversarially Robust Lightweight Neural Architecture.

The adversarial robustness is critical to deep neural networks (DNNs) in deployment. However, the improvement of adversarial robustness often requires compromising with the network size. Existing approaches to addressing this problem mainly focus on the combination of model compression and adversarial training. However, their performance heavily relies on neural architectures, which are typically manual designs with extensive expertise. In this article, we propose a lightweight and robust neural architecture search (LRNAS) method to automatically search for adversarially robust lightweight neural architectures. Specifically, we propose a novel search strategy to quantify contributions of the components in the search space, based on which the beneficial components can be determined. In addition, we further propose an architecture selection method based on a greedy strategy, which can keep the model size while deriving sufficient beneficial components. Owing to these designs in LRNAS, the lightness, the natural accuracy, and the adversarial robustness can be collectively guaranteed to the searched architectures. We conduct extensive experiments on various benchmark datasets against the state of the arts. The experimental results demonstrate that the proposed LRNAS method is superior at finding lightweight neural architectures that are both accurate and adversarially robust under popular adversarial attacks. Moreover, ablation studies are also performed, which reveals the validity of the individual components designed in LRNAS and the component effects in positively deciding the overall performance.

Supporting early-stage design decisions with building performance optimisation: Findings from a design experiment

Building performance optimisation (BPO) has been intensively explored due to its potential in improving building performance and design efficiency. However, studies on its application in supporting early-stage design decisions are rather limited, causing its actual effectiveness to remain questionable. This paper addresses the problem through a design experiment, where participants designed an office building for energy efficiency, using first the conventional method and then BPO. They also evaluated others' design outcomes and completed a questionnaire about the experiment afterwards. The results showed that BPO is effective in supporting early-stage architectural design even for novices. It turned manual designs that are often averagely energy-efficient in the solution space into top designs without significantly affecting other design merits. The average energy use intensity (EUI) decreased by 8.5 % in this experiment. However, it showed a narrow advantage (0.5 % on average) over evaluating a series of design samples and no clear superiority over less-automated supporting methods with more involvement of designers, as its effectiveness is limited by technical difficulties in implementing certain variables strongly correlated with energy efficiency. In addition, while the majority of participants acknowledge the effectiveness of BPO and the acceptance of its design outcomes, manual designs were still preferred by more, mostly for better aesthetics and better alignment of design concepts, implying the necessity of designer's interventions before using BPO outcomes in design practice. The useability of BPO also needs to be improved by shortening the calculation time and lowering the difficulties in building desirable parametric models.

Automated tooth crown design with optimized shape and biomechanics properties

Despite the large demand for dental restoration each year, the design of crown restorations is mainly performed via manual software operation, which is tedious and subjective. Moreover, the current design process lacks biomechanics optimization, leading to localized stress concentration and reduced working life. To tackle these challenges, we develop a fully automated algorithm for crown restoration based on deformable model fitting and biomechanical optimization. From a library of dental oral scans, a conditional shape model (CSM) is constructed to represent the inter-teeth shape correlation. By matching the CSM to the patient’s oral scan, the optimal crown shape is estimated to coincide with the surrounding teeth. Next, the crown is seamlessly integrated into the finish line of preparation via a surface warping step. Finally, porous internal supporting structures of the crown are generated to avoid excessive localized stresses. This algorithm is validated on clinical oral scan data and achieved less than 2 mm mean surface distance as compared to the manual designs of experienced human operators. The mechanical simulation was conducted to prove that the internal supporting structures lead to uniform stress distribution all over the model.

A quantum particle swarm optimization-based optimal LQR-PID controller for load frequency control of an isolated power system

One of the characteristics of a robust power grid is minimal variations in its frequency to load change or loss in generating unit(s). From the perspective of optimal control theory, the issue of load frequency control in the context of the interconnected functioning of power systems is investigated in this work, and a novel load frequency controller is proposed for a single area isolated power network. This novelty incorporates all the primary characteristics of the solutions that are based on a mixture of optimal controller designs by establishing a linear quadratic regulator optimized with quantum particle swarm optimization to design a proportional integral derivative (PID) controller unlike the conventional PID controller designs that are based on a combined Ziegler-Nichols and root locus (ZN-RL) method and manual tuning. The simulation results of the proposed controller using MATLAB show its efficacy in not only ensuring that there is no steady-state error in terms of the system frequency with load changes but also in achieving smoother transients. Following these landmark achievements, a transfer function model of the resulting power grid is constructed. The outcome of the model reveals that the system transients have been improved while keeping the intended steady-state characteristics. Furthermore, it is observed that the proposed load frequency controller has the best performance when compared with the combined ZN-RL method and the manual PID designs. This, therefore, demonstrates the superiority of the proposed design for load frequency control in power systems.

Pelatihan Pembuatan Desain Busana Secara Digital Bagi Peserta Didik SMK Tata Busana di Yogyakarta

This training activity aims to provide knowledge and skills in making digital fashion designs as a continuation of the process of making manual fashion designs for students of SMK Fashion Yogyakarta. The target of the activity is the third-year students of SMK (vocational high school) of the Fashion Expertise Program in Yogyakarta. The lecturers doing off-campus project activities related to fashion design provide training on making fashion designs digitally with the CLO 3D application. In the form of training, the speaker not only delivers material or theory but also conducts training directly to students. This activity is divided into three stages: the preparation stage, the implementation stage, and the evaluation stage. The methods used in this training activity are lectures to convey fashion design theory, demonstration methods to provide examples of the process of making fashion designs digitally, and the practice of making fashion designs using the CLO 3D application carried out by students accompanied by a presenters team. The activity results are (1) Implementing practical activities to make fashion designs with digital techniques attended by three classes, XII Busana 1, 2, and 3, consisting of 90 students. (2) Students are trained in making fashion designs with digital techniques using the CLO 3D application. Based on the evaluation of digitally making fashion designs, participants' attendance during the training reached 96%. While based on student practice results, 62% of student design result is categorized as 'good,' while 38% means fairly good.

Neural Architecture Search with In‐Memory Multiply–Accumulate and In‐Memory Rank Based on Coating Layer Optimized C‐Doped Ge2Sb2Te5 Phase Change Memory

AbstractNeural architecture search (NAS), as a subfield of automated machine learning, can design neural network models with better performance than manual design. However, the energy and time consumptions of conventional software‐based NAS are huge, hindering its development and applications. Herein, 4 Mb phase change memory (PCM) chips are first fabricated that enable two key in‐memory computing operations—in‐memory multiply‐accumulate (MAC) and in‐memory rank for efficient NAS. The impacts of the coating layer material are systematically analyzed for the blade‐type heating electrode on the device uniformity and in turn NAS performance. The random weights in the searched network architecture can be fine‐tuned in the last stage. With 512 × 512 arrays based on 40 nm CMOS process, the PCM‐based NAS has achieved 25–53× smaller model size and better performance than manually designed networks and improved the energy and time efficiency by 4779× and 123×, respectively, compared with NAS running on graphic processing unit (GPU). This work can expand the hardware accelerated in‐memory operators, and significantly extend the applications of in‐memory computing enabled by nonvolatile memory in advanced machine learning tasks.

A scenario-based metaheuristic and optimization framework for cost-effective machine-trail network design in forestry

Designing an optimal machine trail network is a complex locational problem that requires an understanding of different machines’ operations and terrain features as well as the trade-offs between various objectives. With the overall goal to minimize the operational costs of the logging operation, this paper proposes a mathematical optimization model for the trail network design problem and a greedy heuristic method based on different randomized search scenarios aiming to find the optimal location of machine trails —with potential to reduce negative environmental impact. The network is designed so that all trees can be reached and adapted to how the machines can maneuver while considering the terrain elevation’s influence. To examine the effectiveness and practical performance of the heuristic and the optimization model, it was applied in a case study on four harvest units with different topologies and shapes. The computational experiments show that the heuristic can generate solutions that outperform the solutions corresponding to conventional, manual designs within practical time limits for operational planning. Moreover, to highlight certain features of the heuristic and the parameter settings’ effect on its performance, we present an extensive computational sensitivity analysis.

Light weight concrete mix design through fuzzy logic with Cosine similarity aided optimized rules

In this paper, a novel model of mix design built on fuzzy concepts is elaborated. For the development of the model, around 300 lightweight mix design data which is a blend of manually generated data and the data garnered through standard research publications reporting laboratory research and field mix designs have been used. The manual designs are based on ACI guidelines and were generated over spread sheet. have been considered. The model has five inputs viz., density of course aggregate, nominal size of course aggregate, water-cement ratio, target strength and the slump. The outputs are the quantities of cement, water, lightweight aggregate and fine aggregate. This data driven model consisted of four phases, i. Fuzzification of inputs and outputs with appropriate granulation using linguistic terms, ii. Framing of conjunctive rules based on approximate reasoning mapping the antecedent space to consequent space, iii. Selection of similar rules using cosine similarity measure and selection of the rules that generated outputs with minimum error, and iv. testing and validation of the system using error metrics such as root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The so developed model with compact and optimized rule set was tested with test and validation data (15% of the data for each). The results of the system are encouraging with low values of error metrics such as RMSE (10%- 13%), MAE(7–14), and MAPE(7%-14%) over the predicted outputs. The system has also shown high values of coefficient of determination R2 (0.87–0.96) across the predicted outputs. The correlation coefficient (R) authenticating the closeness of predicted values of the outputs with actual values stood in the range of 0.93 – 0.96.

Dapper: Deploying Service Function Chains in the Programmable Data Plane Via Deep Reinforcement Learning

Network functions perform specific packet processing on network traffic. To meet operators' needs, forming service function chains (SFCs) is a fundamental technique used in today's ISPs and datacenter networks. Implementing SFCs in the programmable data plane with high throughput and low latency is a new approach to satisfy demands of ever-growing network traffic. Previous works have proposed different solutions to solve the problem, but they all inevitably have to make trade-offs between running time and performance. For example, an ILP (Integer Linear Programming) can optimize cost but suffers from long running time in large-scale network topologies. Heuristic algorithms depend strongly on manual designs and usually have a performance gap with the optimal solution. In this paper, we propose <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> , a framework for deploying SFCs in the programmable data plane using DRL (Deep Reinforcement Learning) with graph convolutional network. In order to expand the searching space to prevent the optimal value from being missed, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> allows the RL (Reinforcement Learning) agent to simultaneously extract features from both the substrate network and the hardware pipeline, and exploit a graph convolutional network to enhance performance. Moreover, a mask mechanism is also designed to accelerate <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> and improve its scalability. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> has been implemented and extensively evaluated on both P4 hardware switches (equipped with Intel Tofino ASIC) and software switches (i.e., bmv2). Experimental results show that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> can automatically generate deployment solutions in a few seconds of running time after training. They also demonstrate that <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Dapper</i> reduces hardware stage usage and the latency of SFCs by up to 17.8% and 50 <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sim$</tex-math></inline-formula> 73% respectively on average when compared with heuristics.

Stochastic analysis of dig limit optimization using simulated annealing

The results of dig limit delineation in open pit mining are never truly optimized due to gaps in the underlying data, such as insufficient sampling. Aside from the data uncertainty, there is also an influence on the final dig limit by either humans or by the heuristic character of an optimization method like simulated annealing. Several dig limit optimizers have been published, which can replace the manual dig-limits designing process. However, these dig limit designs are generally not adapted to account for this heuristic character. In this paper we present a stochastic analysis tool that can be used with the results of heuristic dig-limit optimization to increase confidence in the obtained results. First, an enhanced simulated annealing algorithm for dig limit optimization is presented. Then, this algorithm is tested on ten different blasts at the Marigold mine, Nevada, USA, as a case study. Finally, the results are analysed with a destination-based ensemble probability map and an analysis conducted of the final solution data distribution. The generated dig-limit designs of the algorithm include high revenue areas that are excluded in comparable manual designs and show improved objective and revenue values. The analysis tool provides block destination probabilities and box plots with the distribution of opportunity value for the dig limit. Furthermore, with the analysis tool, it is possible to make well-informed design decisions in areas of uncertainty.

Interaction-Aware Decision-Making for Automated Vehicles Using Social Value Orientation

Motion control algorithms in the presence of pedestrians are critical for the development of safe and reliable Autonomous Vehicles (AVs). Traditional motion control algorithms rely on manually designed decision-making policies which neglect the mutual interactions between AVs and pedestrians. On the other hand, recent advances in Deep Reinforcement Learning allow for the automatic learning of policies without manual designs. To tackle the problem of decision-making in the presence of pedestrians, the authors introduce a framework based on Social Value Orientation and Deep Reinforcement Learning (DRL) that is capable of generating decision-making policies with different driving styles. The policy is trained using state-of-the-art DRL algorithms in a simulated environment. A novel computationally-efficient pedestrian model that is suitable for DRL training is also introduced. We perform experiments to validate our framework and we conduct a comparative analysis of the policies obtained with two different model-free Deep Reinforcement Learning Algorithms. Simulations results show how the developed model exhibits natural driving behaviours, such as short-stopping, to facilitate the pedestrian's crossing.

Balance-Aware Grid Collage for Small Image Collections.

Grid collages (GClg) of small image collections are popular and useful in many applications, such as personal album management, online photo posting, and graphic design. In this article, we focus on how visual effects influence individual preferences through various arrangements of multiple images under such scenarios. A novel balance-aware metric is proposed to bridge the gap between multi-image joint presentation and visual pleasure. The metric merges psychological achievements into the field of grid collage. To capture user preference, a bonus mechanism related to a user-specified special location in the grid and uniqueness values of the subimages is integrated into the metric. An end-to-end reinforcement learning mechanism empowers the model without tedious manual annotations. Experiments demonstrate that our metric can evaluate the GClg visual balance in line with human subjective perception, and the model can generate visually pleasant GClg results, which is comparable to manual designs.

Experiment on the Flexural Functioning of Cold-Formed Steel Built-Up Complex Hat Section

Abstract: Cold-formed steel members are comprehensively utilized in the building construction industry, especially in residential, commercial, and industrial buildings. Thin sheet steel products are widely noticed in the building industry and range from purlins to roof sheeting and floor decking. Generally, these are used for basic building elements for the congregation at the site or as prefabricated frames or panels. They obtained the generic title ‘ Cold-Formed Steel Sections. The uses of these products are many and varied, ranging from “tin” cans to structural piling, from keyboard switches to mainframe building members. This paper dispenses an experimental and software analysis of the flexural behavior of cold-formed steel built-up sections. The experimental results are also verified with finite element analysis using Manual Designs and ANSYS Software. The analytical results obtained are better exposure to the experimental results.