17315 research outputs found
Sort by
A biologically inspired filter significance assessment method for model explanation
Get PDFNeuron importance assessment is crucial for understanding the inner workings of artificial neural networks (ANNs) and improving their interpretability and efficiency. This paper introduces a novel approach to neuron significance assessment inspired by frequency tagging, a technique from neuroscience. By applying sinusoidal contrast modulation to image inputs and analyzing resulting neuron activations, this method enables finegrained analysis of a network’s decision making processes. Experiments conducted with a convolutional neural network for image classification reveal notable harmonics and intermodulations in neuron-specific responses under part based frequency tagging. These findings suggest that ANNs exhibit behavior akin to biological brains in tuning to flickering frequencies, there by opening avenues for neuron/filter importance assessment through frequency tagging. The proposed method holds promise for applications in network pruning, and model interpretability, contributing to the advancement of explainable artificial intelligence and addressing the lack of transparency in neural networks. Future research directions include developing novel loss functions to encourage biologically plausible behavior in ANNs
A UV-free and cytocompatible crosslinking strategy for methacrylated chitosan hydrogels via bisulfite-initiated polymerization for embedded 3D bioprinting
No full textMethacrylated chitosan (ChiMA) is a versatile polymer ink for fabricating 3D tissue-like structures due to its biocompatibility and processability. However, traditional UV-based crosslinking methods offer limited precision control of light intensity and pose risks for cell-laden hydrogels. This study introduces a novel, UV-free crosslinking strategy based on bisulfite-initiated radical polymerization for ChiMA hydrogels. By integrating bisulfite-initiated radical polymerization into an embedded printing process, ChiMA ink was successfully printed within a sodium bisulfite (SBS)-containing support bath to fabricate complex-shaped 3D structures. ChiMA hydrogels crosslinked with SBS were characterized for structural, mechanical, rheological, morphological, swelling, degradation, and cytocompatibility properties. Their mechanical properties increased notably with higher SBS content, as Young's modulus ranged from 20.5 to 151.3 kPa and compressive strength from 14 to 80 kPa. Gelation of the ChiMA solution, occurring within seconds at neutral pH, was delayed to 4–23 min in the slightly acidic environment. Swelling ratios decreased from 36 % to 29 %, and average pore size from 220 μm to 105 μm with increasing SBS. The hydrogels with 0.05 % SBS degraded by ∼20 %, while those with higher SBS (0.25–0.5 %) degraded only ∼5 % over 28 days. Cell viability remained consistently high (>97 %) throughout the 7-day culture period in all bioprinted samples
Data file for multi-commodity vehicle routing problem with pickup and delivery for electric micro-mobility devices rebalancing and battery swapping
No full textThis study addresses a variant of the multi-commodity vehicle routing problem that arises in electric micro-mobility systems, where a fleet of capacitated vehicles is tasked with simultaneously performing device rebalancing and battery swapping operations. The objective is to satisfy station-level demands while minimizing total travel cost. We refer to this problem as the Electric Micro-Mobility Device Rebalancing and Battery Swapping Problem (EMDRBS), which extends the classical pickup and delivery vehicle routing model by incorporating multi-commodity flows and battery replacement constraints. To solve EMDRBS, we propose four mixed-integer linear programming (MIP) formulations and evaluate their computational performance on a set of realistic benchmark instances derived from public bike-sharing data. While two formulations exhibit strong performance on medium- and large-scale instances, solving the largest cases to optimality remains computationally challenging. To address it, we develop a fix-and-optimize matheuristic that dynamically adjusts its destruction strategies based on past performance and repairs partial solutions through restricted MIP reoptimization. Computational results show that among the four MIP formulations, F3 and F4 provide the best trade-off between strength and scalability, while the proposed FixOpt matheuristic delivers comparable solution quality with substantially shorter runtimes on large instances. Moreover, coordinated planning of rebalancing and battery swapping reduces total travel distance by up to 50\%—particularly under high demand and limited vehicle capacity—without increasing depot resource requirements, demonstrating the practical value of integrated optimization
If it ain't broke, should you still fix it? Effects of incorporating user feedback in product development on mobile application ratings
No full textHow can firms utilize collective user feedback in reviews to better tailor their products to improve customer satisfaction? Based on an automated text analysis of 1,075,704 reviews and a content analysis of 3255 mobile application updates, observed over 460 apps’ first year on the market, this paper investigates the role of collective user feedback on user ratings during the process of developing successive mobile app generations. The results reveal that the rewards associated with responding to user feedback and the penalties due to ignoring this feedback can be substantial. The impact of the match/mismatch between user feedback and product development decisions depends on the topic of the feedback and the timing of the update. The findings provide app developers with guidance on the challenge of prioritizing possible development paths: (1) improve ratings by promptly matching user feedback that require new content or smooth functioning of the app; (2) avoid rating penalties by continuously improving existing content or ensuring compatibility with most recent operating systems and devices. The implications shed light on the fundamental question of whether and when to pro-act on or re-act to the voice of the customer
Off-target structural insights: ArnA and AcrB in bacterial membrane-protein cryo-EM analysis
No full textMembrane-protein quality control in Escherichia coli involves coordinated actions of the AAA+ protease FtsH, the insertase YidC and the regulatory complex HflKC. These systems maintain proteostasis by facilitating membrane-protein insertion, folding and degradation. To gain structural insights into a putative complex formed by FtsH and YidC, we performed single-particle cryogenic electron microscopy on detergent-solubilized membrane samples, from which FtsH and YidC were purified using Ni–NTA affinity and size-exclusion chromatography. Although SDS–PAGE analysis indicated high purity of these proteins, cryo-EM data sets unexpectedly yielded high-resolution structures of ArnA and AcrB at 4.0 and 2.9 Å resolution, respectively. ArnA is a bifunctional enzyme involved in lipid A modification and polymyxin resistance, while AcrB is a multidrug efflux transporter of the AcrAB–TolC system. ArnA and AcrB, known Ni–NTA purification contaminants, were also consistently detected by mass spectrometry in Strep-Tactin affinity-purified samples, validating their presence independently of affinity-tag selection. ArnA, which is typically cytoplasmic, was consistently found in membrane-isolated samples, indicating an association with membrane components. Only 2D class averages corresponding to the cytoplasmic AAA+ domain of FtsH were observed; neither side views of full-length FtsH nor densities corresponding to an intact FtsH–YidC complex could be identified, due to the conformational flexibility of the FtsH complex and its transient interaction with YidC, which limited particle alignment and stable classification in cryo-EM data sets. Two-dimensional class averages revealed additional particles resembling GroEL and cytochrome bo3 oxidase. These results underscore the utility of cryo-EM in uncovering off-target yet structurally well defined complexes, which may reflect physiologically relevant interactions or purification biases during membrane-protein overexpression
The Directional Factor As A Key Determinant Of Chatter-Free Robotic Milling In Light Alloys
Get PDFRobotic machining can be a cost-effective alternative to conventional CNC systems, especially for processing large or complex workpieces. However, the inherently low stiffness and posture-dependent dynamics of industrial robots results in them being highly susceptible to chatter, especially at low frequencies. Thus, this thesis focuses on the identification of low-frequency chatter in robotic milling, emphasizing the role of structural dynamics and the mean directional factor (MDF) in chatter formation. This is novel because there are some studies that attribute low-frequency chatter to mode-coupling in literature. To study the effect of MDF in a controlled environment, a custom flexure-based fixture was designed to mimic the dominant low-frequency mode typically observed in robotic arms. This fixture offers a single-mode-dominant system, which provides repeatable testing and clear interpretation of dynamic behaviour. Modal analysis was carried out using finite element modelling (FEM) and validated through experimental modal analysis (EMA). High correlation between simulation and test results confirmed the accuracy of the simplified model. The system was finally tested under machining conditions, validating theoretical predictions related to mean directional factors (MDF) and demonstrating that slotting operations can effectively reduce chatter. The results confirm that positive MDF values are associated with stable cutting, while negative MDF values lead to low-frequency self-excitation. This study offers a practical framework for enhancing stability in robotic milling and contributes to the development of chatter-resistant robotic systems for high-speed machinin
Method Of Weighted Words On Cylindric Partitions
Get PDFIn this thesis, we study the generating functions of cylindric partitions having profilec = (c1, c2, . . . , cr) with rank 2 and levels 2,3 and 4. As a result, we give expressionsalternative to Borodin’s formula for these generating functions. We use the methodof weighted words which was first introduced by Alladi and Gordon, later was appliedby Dousse in a new version to prove some partition identities and to get infiniteproducts. We adapt the method to our subject with a more combinatorial approach
Domain Generalized Remote Sensing Scene Captionıng Via Country-Level Geographic Information
Get PDFThis thesis investigates the impact of incorporating country-level, text-based geographicalinformation into a large-scale vision-language model fine-tuned for captioningoptical remote sensing imagery. We hypothesize that enriching visual inputs withcorresponding geographical context can enhance model performance, particularly ingeneralizing to images from previously unseen countries or continents. To test this,we fine-tune the Large Language and Vision Assistant (LLaVA) on optical satelliteimages from European countries, augmenting them with textual geographicaldescriptions, and evaluate its performance on images from other global regions. Experimentsconducted across 175 countries using the newly released SkyScript datasetreveal that even lightweight geographical context—extracted from Wikipedia—canmitigate cross-country domain shifts, leading to notable improvements in captioningaccuracy. These findings highlight the potential of multimodal approaches inenhancing the geographic generalization capabilities of vision-language models
Is process damping effective in the stability of robotic milling?
No full textChatter stability is a major constraint in milling, where low and high cutting speeds are used. At low cutting speed regime, process damping leads to increased stability, whereas at high cutting speeds lobing effect is beneficial. Excitation frequency depends on spindle speed and the number of cutting flutes on the milling tool. Hence the vibration mode governing chatter stability varies for multi-mode milling systems. In CNC milling, low frequency structural modes are stiffer than cutting spindle-holder-tool (SHT) assembly. However, robotic milling demonstrates a distinct behavior as low frequency modes are significantly more flexible. This study investigates the effect of robot structure induced low frequency vibration modes on stability limits at low cutting speeds, where process damping is expected to increase stability limits. Time domain simulations are used to explain the variation of the dominant mode from high frequency to low frequency with the decreasing spindle speed. Simulated stability diagram for the multi-mode robotic milling system is verified by experiments. It was shown that especially the vibration modes in the range of 15 to 20 Hz do not generate enough process damping force due to long vibration waves, i.e. cutting speed – to – chatter frequency ratio, when low frequency modes govern chatter stability. Simulation of stability diagrams showed that there is a spindle speed region where the stability lobes governed by the robot structure crosscut the stability lobes governed by the THS assembly. Due to the inherent effect of tool diameter (D) and number of cutting flutes (Z) on cutting speed and excitation frequency, this region shifts according to the D/Z ratio. It was shown through simulations that D/Z ratio is a critical metric to benefit from process damping without the interference of the low frequency excitation of the robotic structure. The simulation results are used to provide suggestions for milling tool selection in robotic milling, where the main conclusion is to use lower D/Z ratio, which means that using high number of cutting flutes