GIST Scholar
Not a member yet
30271 research outputs found
Sort by
Solution Augmentation for ARC-AGI Problems Using GFlowNet: A Probabilistic Exploration Approach
No full textOne of the core challenges in building general reasoning systems lies in generating diverse, human-aligned solution trajectories—different yet valid paths by which a problem can be solved. Prior approaches often rely on handcrafted templates, rule-based augmentations, or human demonstrations, which are limited in scalability and stylistic diversity. To address this, we explore the use of Generative Flow Networks (GFlowNets) for automated solution augmentation in reasoning tasks. We propose a framework that learns to generate diverse reasoning trajectories with probabilities proportional to their quality, guided by a human-inspired reward function and a novel geometric forward policy. This enables the generation of multiple plausible solution paths without relying on manual supervision. Moreover, our method supports efficient test-time augmentation from input-output examples alone, without access to ground-truth programs or external demonstrations—making it suitable for zero-shot settings. We evaluate our framework on the Abstraction and Reasoning Corpus (ARC-AGI), a benchmark designed to test compositional and abstract reasoning. Our results show that GFlowNets can effectively explore the space of valid reasoning processes, producing a variety of plausible reasoning trajectories, similar to how different individuals might solve the same problem using different intermediate steps. These trajectories are generated at scale—over 100k per task in under an hour, and follow a logarithmic yield trend, enabling practical tradeoffs between augmentation volume and novelty. Furthermore, fine-tuning a large language model (LLaMA 3.1 Instruct 8B) on these synthetic trajectories leads to a 28.6% improvement in reasoning accuracy on ARC tasks, demonstrating the downstream utility of our method. These findings suggest that GFlowNets offer a promising foundation for modeling structured reasoning in automated trajectory generation. Our code is here: https://github.com/GIST-DSLab/GFN_to_ARC.TRUEforeig
Enhanced photoelectrochemical glycerol oxidation and hydrogen production using Fe-doped ZnS/ZnO nanowire heterostructures
No full textAchieving efficient photoelectrochemical (PEC) glycerol oxidation requires understanding and optimizing interactions between glycerol and the photoelectrode surface. In this study, we present Fe-doped ZnS/ZnO photoelectrodes, focusing on their surface properties. Incorporating Fe into ZnS improves photocatalytic performance, promoting glycerol oxidation and hydrogen production to nearly twice that of ZnO. At 1.0 V versus the reversible hydrogen electrode (VRHE), the Fe-doped ZnS/ZnO nanowires achieve a photocurrent density of 1.88 mA cm−2 and a total liquid product yield of 177.7 mmol m−2 h−1. Glyceric acid is identified as the major oxidation product, with sulfur-based structures favoring C3 product formation. Computational results also reveal that Fe-doping on the surface of ZnS can facilitate the adsorption of hydroxyl groups of glycerol, leading to efficient photocatalytic reactions on ZnS. This study clarifies the role of metal dopants in enhancing PEC glycerol oxidation and provides valuable insights for designing reactant-tailored photoelectrodes. © 2025FALSEsciescopu
Dataset Distillation for Super-Resolution Without Class Labels and Pre-Trained Models
No full textTraining deep neural networks has become increasingly demanding, requiring large datasets and significant computational resources, especially as model complexity advances. Data distillation methods, which aim to improve data efficiency, have emerged as promising solutions to this challenge. In the field of single image super-resolution (SISR), the reliance on large training datasets highlights the importance of these techniques. Recently, a generative adversarial network (GAN) inversion-based data distillation framework for SR was proposed, showing potential for better data utilization. However, the current method depends heavily on pre-trained SR networks and class-specific information, limiting its generalizability and applicability. To address these issues, we introduce a new data distillation approach for image SR that does not need class labels or pre-trained SR models. In particular, we first extract high-gradient patches and categorize images based on CLIP features, then fine-tune a diffusion model on the selected patches to learn their distribution and synthesize distilled training images. Experimental results show that our method achieves state-of-the-art performance while using significantly less training data and requiring less computational time. Specifically, when we train a baseline Transformer model for SR with only 0.68% of the original dataset, the performance drop is just 0.3 dB. In this case, diffusion model fine-tuning takes 4 hours, and SR model training completes within 1 h, much shorter than the 11-hour training time with the full dataset.FALSEsciescopu
Mesenchymal stem cell-laden double-network hydrogel nerve guidance conduits for peripheral nerve injury repair
No full textTo enhance the repair of peripheral nerve injuries (PNIs), various nerve guidance conduits (NGCs) have been developed by integrating topological, biochemical, and cellular cues. Hydrogel-based NGCs are particularly promising owing to their unique tissue-mimicking characteristics, such as high water content, softness, and porosity. However, their weak mechanical strength and insufficient biological activity limits their application. Therefore, in this study, we aimed to develop NGCs by encapsulating human umbilical cord-derived mesenchymal stem cells (ucMSCs) in double-network (DN) hydrogel conduits for improved peripheral nerve regeneration. A DN hydrogel, fabricated via sequential photo- and ionic-crosslinking of 15 % gelatin methacrylate and 1 % alginate, exhibited excellent rheological and mechanical properties, including fatigue resistance, suture retention, and kink resistance. In a rat sciatic defect model, ucMSC-encapsulated DN NGCs demonstrated significantly improved functional and structural recovery compared to medical silicone and non-cellular hydrogel NGCs. Quantitative assessments revealed that the MSC-laden NGC group exhibited superior functional recovery, as indicated by footprint analysis, electromyography, thermal withdrawal latency, and muscle weight restoration. Moreover, histological analysis and transmission electron microscopy confirmed significantly enhanced axonal regeneration and myelination in the MSC-laden NGC group (axon diameter and myelin thickness). Overall, our results indicate that the MSC-laden hydrogel NGCs can serve as a novel platform to treat PNIs and function as effective stem cell delivery scaffolds for the regeneration of various tissues, such as the skin, tendons, and muscles.TRUEsciescopu
Decoding the evolution and dynamics of semicrystalline block copolymer assembly via liquid-phase transmission electron microscopy
No full textNature utilizes self-assembly to form complex, functional structures, inspiring advanced materials design. Polymer crystallization drives assemblies with both ordered and disordered regions. Crystallization-driven assembly of BCPs enables unique hierarchical nanostructures with enhanced colloidal stability and directionality, applicable from optoelectronics to biomedicine. However, mechanisms governing morphological transitions remain poorly understood due to complex microphase separation and competitive crystallization. Using liquid-phase transmission electron microscopy, we visualize the spontaneous assembly of semicrystalline amphiphilic BCPs. We observe structural transformations from unimers to spherical, cylindrical, toroidal micelles, and vesicles by varying constituent block ratios. Image segmentation overcomes low contrast of aqueous assemblies, enabling motion tracking. Nanostructures exhibit structural evolution driven by long-range hydrophobic interactions from formed elemental micelles undergoing anomalous diffusion. Notably, toroid formation follows a distinct pathway compared with conventional BCPs due to semicrystalline BCPs’ preference for low curvature at the core-corona interface. Insights into assembly dynamics via real-time imaging provide strategies for controlling complex hierarchical structures. © 2025 Elsevier Inc.FALSEsciescopu
Challenge and opportunity in scaling-up hydrogen production via electrochemical ammonia electrolysis process
No full textHydrogen as a clean carrier has received considerable attention due to its potential in sustainable energy systems. However, the prevalent hydrogen production methods, notably steam reforming from natural gas, present environmental challenges, primarily CO2 emissions. In here, we aim to provide insights into the scalability of hydrogen production through electrochemical ammonia electrolysis to hydrogen production (eAEH), proposing ammonia as an effective hydrogen carrier to mitigate these environmental concerns. Our focus is on the ammonia oxidation reaction (AOR) within the electrochemical decomposition framework, underscoring the operation at a low reversible cell voltage to significantly enhance energy efficiency. We explore strategies to scale up eAEH, analyzing the potential and limitations of various electrocatalysts, and examining the feasibility of employing machine learning techniques for optimal catalyst selection. Thus, the AOR research represents a pivotal technological innovation for the sustainable energy transition, potentially establishing a critical foundation for advancing a hydrogen economy. © 2024 Elsevier B.V.FALSEsciescopu
Smart Fault Detection in Electric Vehicles Using Battery and Motor Operation Data Driven Deep Learning
No full textIdentification of cell-type-specific, transcriptionally active transposable elements using long-read RNA-sequencing data-based comprehensive annotation
No full textBackground: The biological functions of transposable element (TE)-derived transcripts during physiological development, disease development, and progression have been previously reported. However, research on locus-specific TE-derived transcript expression in various human cell types remains limited. Methods: We processed 2596 publicly available human long-read RNA-sequencing (LR RNA-seq) datasets covering 21 organs and 71 cell lines in both healthy individuals and diseased patients with various conditions to compile this TE-derived transcript annotation. We established a pipeline for assembling transcripts containing TE sequences to measure transcriptionally active TE-derived transcripts in diverse tissues and cell types. Next, we applied our TE annotation to the Genotype-Tissue Expression (GTEx) single-cell RNA-sequencing (scRNA-seq) data from eight tissues. Results: We constructed the first transcriptom6e-based TE annotation using massive amounts of human LR RNA-seq data for use as a comprehensive reference to detect locus-specific TE-derived transcripts. Our annotation showed better detection accuracy for TE-derived transcripts than the RepeatMasker and GENCODE nonTE gene annotations. This annotation enabled the identification of novel TE-derived transcripts and their isoforms. We also identified alternative transcription end sites for long noncoding genes and confirmed previously annotated TE-nonTE gene fusion transcripts. Next, we applied our TE-derived transcript annotation to public scRNA-seq data from various human tissues and identified several cell-type-specific TE-derived transcripts in a locus-specific manner. Conclusions: We generated a comprehensive, TE-derived transcript annotation using large-scale, LR RNA-seq data. Researchers can use our TE reference annotation to analyze active TE transcripts and their splicing isoforms in specific transcriptome datasets and to detect de novo TE transcripts. The discovery of cell-type-specific TE-derived transcripts may help explain mechanisms underlying the maintenance of cellular identity and provide new insights into the pathological mechanisms of various diseases. © 2025 Elsevier B.V., All rights reserved.TRUEscopu
Phonon Mean Free Path Spectrum of Single Crystalline Silicon Thin Film
No full textWith the advanced fabrication technology, the density of transistor keeps increasing and the design is getting more complexed. Fourier’s law, which is applied for heat conduction in bulk materials, becomes unsuitable for thermal transport at this micro- and nano-scale system. Especially, when phonon mean free path (MFP) is longer than the characteristic length of a system, phonons in the system tends to be scattered significantly at the boundary, leading to the reduction of thermal conductivity. Therefore, nanostructuring is one of attractive ways to manipulate the thermal conductivity of material in some applications such as modeling thermal management and designing thermoelectric materials. To predict the extent of thermal conductivity reduction via nanostructuring, it is significantly crucial to figure out the contribution of each MFP to thermal conductivity, which is knowon as phonon MFP spectrum. Thus, there have been numerous theoretical and experimental studies up to date. The 1st principle calculation has been successful to predict the MFP spectrum for simple atomic structures such as Si, Ge, and etc. However, it is challenging and time consuming to extend the same approach to nanostructures. Among experimental approaches, the TDTR (time-domain thermoreflectance) and TTG (thermal grating method) are representative for this research but have some limitations. The former adopts a metal transducer, which makes the analysis complicated, and the latter is limited in decreasing the size of grating, i.e., the characteristic length, due to the diffraction limit of light. Also, most of previous works have studied only about bulk-scale materials, even though modern micro- or nanoelectronic devices are often based on thin film structures. In this work, nanoslits with various widths, which were patterned on a suspended silicon nano-film using EBL (e-beam lithography) and RIE (reactive ion etching process), provided varied ballistic thermal resistances. The effective thermal conductivity values of nanoslit-films were measured in a temperature range of 40–300 K using micro-suspended devices, which were individually fabricated from SOI (silicon-on-insulator) wafer. Additionally, to better understand the measurement results, discrete ordinate method was used with inputs by 1st principle calculation. From the measurement results, the phonon MFP spectrum was extracted by a convex optimization, using suppression function obtained by solving freqeuency-independent Boltzmann transport equation. This study offers, for the first time, the phonon MFP spectrum for Si thin film using a non-optical experimental approach.DoctorAbstract ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ i
Contents ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ iii
List of Figures ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ vi
List of Tables ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ vii
CHAPTER 1. INTRODUCTION․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 1
1.1. Phonon: carrier of heat energy․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 1
1.2. Importance of phonon MFP․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 2
1.3. Phonon MFP spectrum ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 2
1.4. Previous experimental studies․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 3
1.5. Organization of this Thesis․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 4
CHAPTER 2. EXPERIMENTAL METHOD ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 5
2.1. Introduction ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 5
2.2. Evolution of measurement device design․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 5
2.3. Fabrication of nanoslit pattern․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 9
2.4. Measurement device preparation․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 11
2.5. AC heating measurement method․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 13
2.6. Third and second harmonic analysis for measurement validation․․․․․․․․․․․․․․․․․․․․․․․ 14
2.7. TCR fitting and compensation methodology․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 15
2.8. Experimental setup․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 18
2.9. Summary․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 19
CHAPTER 3. NUMERICAL METHOD ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 20
3.1. Introduction ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 20
3.2. Workflow of the numerical method․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 20
3.3. Phonon properties from DFT calculation․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 21
3.4. Discrete ordinate method for phonon Boltzmann transport equation․․․․․․․․․․․․․․․․․․․․ 22
3.4.1. Integrating phonon properties․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 22
3.4.2. Phonon Boltzmann transport equation with phonon intensity․․․․․․․․․․․․․․․․․․ 23
3.4.3. Boundary condition․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 25
3.5. Verification of model reliability and convergence testing․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 26
3.6. Summary․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 28
CHAPTER 4. MEAN FREE PATH SPECTRUM EXTRACTION PROCESS․․․․․․․․․․․․․․․․․ 28
4.1. Introduction․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 28
4.2. Phonon MFP spectrum extraction․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 29
4.3. Suppression function ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 30
4.4. L-curve method ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 32
4.5. Summary․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 33
CHAPTER 5. RESULTS and DSICUSSION․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․.․․․․․․․․․․․․․․․. 34
5.1. Thermal conductivity measurement of Si nano-film․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 34
5.2. Verification of ballistic thermal transport in experimental results․․․․․․․․․․․․․․․․․․․․․․․․ 35
5.3. Verification of ballistic thermal transport in numerical results․․․․․․․․․․․․․․․․․․․․․․․․․․․ 36
5.4. Phonon MFP spectra of single crystalline Si thin film․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 37
CHAPTER 6. SUMMARY AND FUTURE WORKS․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 39
6.1. Summary․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 39
6.2. Future works․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 40
References․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․.․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․.․ 41
Acknowledgements ․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․.․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․․ 4
A 56-Gb/s DAC/ADC-Based Multicarrier Transceiver With TX Polar DSP and RX MIMO-DSP for >40-dB Loss Channel
No full textThis letter presents a digital-to-analog-converter/analog-to-digital converter (DAC/ADC)-based multicarrier transceiver fabricated in a 22-nm FinFET technology. The multicarrier signaling scheme utilizes orthogonally spaced carriers for spectral efficient band spacing and exhibit jitter robustness compared to conventional baseband pulse amplitude-based signaling. The transmitter has a polar digital signal processor (DSP) to generate the equalized codes driving the 7-b phase DACs, and 7-b amplitude DACs with 2-b predistortion to yield 1.2-Vppd swing. The multicarrier receiver front-end ADC outputs are equalized with a MIMO DSP backend to compensate for the intersymbol interference and interchannel interference. The measured transceiver at 56 Gb/s through a 40.8-dB loss at 14-GHz channel showed a jitter tolerance of up to 1.21 psrms at BER <10-4 with 7.82-pJ/bit power efficiency. © 2018 IEEE.FALSEscopu