text_with_holes stringlengths 136 3.39k | text_candidates stringlengths 49 1.31k | A stringclasses 6
values | B stringclasses 6
values | C stringclasses 6
values | D stringclasses 6
values | label stringclasses 4
values |
|---|---|---|---|---|---|---|
<|MaskedSetence|> <|MaskedSetence|> For MCTS, we solve for two different sets of weights, one assigning equal importance (adjusted by the objective units) which we call MCTS Base and one in which we emphasize renewable energy deployment (MCTS RE). The value iteration solver uses a discretized version of the state spa... | **A**: We compare these policies based on final weighted objectives such as remaining budget, equity in energy access, and renewable energy penetration.
**B**:
We compare the MDP optimal policy with several benchmark policies: random policy (Random), expert heuristic (Expert), and MDP-optimized policy.
**C**: We com... | BCA | BCA | ABC | BCA | Selection 1 |
<|MaskedSetence|> A key future direction is to implement and evaluate CERA in online simulations. While this study focuses on moist-physics parameterization, the framework is broadly applicable and could be extended to other parameterized processes, or even to applications beyond parameterization. <|MaskedSetence|> W... | **A**: This imposes inherent limits on the generalizability of this approach.
**B**: Notably, CERA does not degrade performance in the control climate and in some cases even improves it, suggesting that latent alignment can enhance generalization without sacrificing in-distribution accuracy.
**C**: For example, CERA’... | BCA | BCA | BCA | BAC | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> The latent space encodes the 3D ocean space using two spatial dimensions, while the third dimension captures the temporal sequence of ocean states. The input consists of a sequence of four consecutive daily mean values (Xt−3,Xt−2,Xt−1,XtX_{t-3},X_{t-2},X_{t-1},X_{t}) of key 3D ocea... | **A**: Each block contains a sequence of Shifted Windows Attention (Swin) 3D modules, where each module comprises a multi-head attention layer, a normalization layer, a multi layer perceptron (MLP), and an additional normalization layer.
**B**: 2 Model
The MedFormer architecture is based on a U-Net structure, with bo... | BAC | BAC | CBA | BAC | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> And then, in the quantization phase, the CSI extracted by Alice and Bob is quantized into the raw bit sequences. Next, during the information reconciliation, the error correction codes are used to correct discrepancies in the raw bit sequences between Alice and Bob. Finally, the pr... | **A**: The basic process of PLKG consists of four main steps: channel probing, quantization, information reconciliation, and privacy amplification [5].
**B**: During the channel probing, Alice and Bob transmit the pilot signal to share the CSI in a channel coherence time.
**C**: Although the effectiveness of PLKG has... | ABC | ABC | ABC | CAB | Selection 3 |
6 Conclusion
Overall, in this study, we developed a number of machine learning models, including SVMs, LSTMs, and CNNs, for inferring emotions from human speeches. <|MaskedSetence|> <|MaskedSetence|> This is a promising result, given the small size of our training set. <|MaskedSetence|> In addition, we demonstrate... | **A**: Our best model was a ResNet34 neural network, which achieved an accuracy of 66.7%66.7\% and an F1 score of 0.6310.631.
**B**: With more training data, the model will definitely be able to learn better and recognize emotion classes with higher accuracy levels.
**C**: Our models were trained and evaluated on sma... | CAB | CAB | CAB | ACB | Selection 1 |
Data Integration: The database received through the KIRETT project consisted of 83 CSV files, each containing information about complete rescue events, while some records had multiple rows on the same patient to reflect how patient’s health is evolving with time. <|MaskedSetence|> The Python program went through each... | **A**: If the information is the same in different records, then the particular cell contains only one data point.
**B**: Otherwise, all different data are stored with a comma separator.
**C**: Therefore, to further process the data, these files and patient information were merged using a Python program to obtain a h... | CAB | CAB | ACB | CAB | Selection 2 |
<|MaskedSetence|> One of the participants interviewed is a non-expert, and one is a professional videographer. <|MaskedSetence|> A recurring critique was that the videos lacked a unifying artistic vision, with inconsistent colour palettes, lighting, and overall visual styles from one shot to the next. This disjointed... | **A**: This issue was compounded by a lack of character consistency, where protagonists would frequently change appearance or even species (”Vocal Jazz, LALM-Based method” example555https://github.com/goodPointP/Results-For-Music-Visualization-Generation-Pipeline?tab=readme-ov-file#vocal-jazz-1) between shots, breaking... | BCA | CBA | BCA | BCA | Selection 1 |
<|MaskedSetence|> For XGBoost and CatBoost, we use the default parameters provided by the Python libraries of [2, 3].
Neural network: For the neural network models, we use the Adam optimizer [17] with a learning rate of 1×10−31\times 10^{-3} and batch size of 32. The training process continues for 100k epochs. The p... | **A**: The output layer generates predictions for all VminV_{min} patterns simultaneously.
.
**B**: The embedding layer maps the fused features into a 32-dimensional vector.
**C**: Linear regression, XGBoost, and CatBoost: Following [10, 4, 15], we apply Correlation Feature Selection (CFS) [16] to select the three ... | CBA | CBA | CBA | ACB | Selection 1 |
Figure 1: Migration performance of 12 knowledge distillation methods (Cumstomkd(Lee et al. 2025), FitNet(Romero et al. <|MaskedSetence|> 2019), SP(Tung and Mori 2019), FNKD(Xu et al. <|MaskedSetence|> 2022), Logit(Sun et al. 2024), GA(Wang et al. 2018), Vlite(Jang, Ma, and Lee 2025), DHO(Kang et al. 2025),AMMKD) on ... | **A**: 2014), RKD(Park et al.
**B**: This training is performed via the contrast loss method to achieve uniform embedding space learning of multimodal signals, thus efficiently aligning the representations of different modalities.
**C**: 2020), KD(Hinton 2015), DKD(Zhao et al.
| ACB | BCA | ACB | ACB | Selection 1 |
Multi-axis 3D printing has emerged as a rapidly expanding research area due to the advantages offered by high DoF in motion, which enable material deposition along wireframes of lightweight structures (Huang et al., 2016; Wu et al., 2016), curved layers within solid volumes (Dai et al., 2018), and shell structures (Mi... | **A**: These capabilities are often achieved by computing scalar fields whose isosurfaces satisfy both design and manufacturability requirements, and are then extracted as working surfaces for printing.
Such printing layers have also been generated through optimized spatial deformation, forming layers that are either... | ABC | ACB | ACB | ACB | Selection 3 |
3.3 Emanation from USB Cable
To test emanation from USB cables, a USB mouse was used. It was USB 2.0, which supports a 12 Mbps data rate at full speed. <|MaskedSetence|> The strongest emanation was found at the 4th4^{th} harmonic (48 MHz), which is shown in Fig. 2(d). <|MaskedSetence|> <|MaskedSetence|> The sold... | **A**: Fig. 2(e) shows the maximum emanation power versus distance plot for up to 160 cm for USB cables with three types of repairing process (twisting, soldering, and using butt connector).
**B**: Hence, 12 MHz and its harmonics were our primary targets.
**C**: Data were collected from the nearest position to the ca... | BCA | BAC | BCA | BCA | Selection 1 |
<|MaskedSetence|> (2020) on model performance and its ability to address the catastrophic forgetting problem. <|MaskedSetence|> The results show that combining SI (Zenke et al. <|MaskedSetence|> However, this combination also led to performance degradation during the training process (2(right)). Moreover, 4 shows th... | **A**:
In this report, we investigate the impact of the brain-inspired replay proposed by Van de Ven et al.
**B**: Our analysis primarily focuses on internal replay, identified as the most critical component of the model.
**C**: (2017)) with the BIR model, as well as the internal replay itself, helps mitigate catast... | CAB | ABC | ABC | ABC | Selection 4 |
The findings of this study uncover the potential and constraints of Advanced Driver Assistance Systems (ADAS) in reducing pedestrian fatalities. <|MaskedSetence|> Evidence-based research shows that the automobiles equipped with ADAS are responsible for fewer high-speed pedestrian fatalities than automobiles without A... | **A**: But the effectiveness of ADAS varies in all instances, and the study identifies certain conditions in which the impact is severely reduced [13].
**B**: Key results from the efficacy of ADAS in pedestrian safety indicate that certain ADAS components, particularly Pedestrian Automatic Emergency Braking (PAEB) and... | BAC | BAC | BCA | BAC | Selection 1 |
Figure 4: Analysis of classical isotherm model performance, optimization methodology, and physics validation across geological lithologies. (a) Lithology-specific model performance matrix displaying R2 values for five classical isotherm models across three primary geological formations. Clay minerals demonstrate opti... | **A**: (d) Comparative analysis of PyTorch-based automatic differentiation versus traditional least-squares optimization methodologies across four critical performance metrics.
**B**: (b) Systematic residual analysis across pressure regimes indicates consistent under-prediction patterns at intermediate pressures (1–10... | CAB | BCA | BCA | BCA | Selection 2 |
Experimental Design Challenges : Aside from participant expertise, design of a listening test is not standardized as well with factors to consider like- sample selection, environment setting of the listening test and phrasing of the surveys. Environment variations, confusing phrasing of the surveys and small sample siz... | **A**: (2024) defined musicality as how much the given sound is melodiousness and harmoniousness, whereas Yuan et al.
**B**: For example, in their listening test, Schneider et al.
**C**: (2024) defined musicality based on two aspects- the overall consistency of the music in terms of melodic patterns and chord progres... | BAC | BAC | ACB | BAC | Selection 2 |
To achieve flexible task adaptation, we propose to model the task-specified information in the task prompt instead of training task heads. However, both the reasoning steps and used information of different tasks vary significantly. <|MaskedSetence|> This module aims to address this challenge with a few labeled seed t... | **A**: Task Prompt Optimization.
.
**B**: Next, we first introduce the construction of task prompt template, and then describe the optimization process.
Figure 3.
**C**: It is challenging to construct a data-invariant task prompt that can be easily understood by MLLMs and suitable for different trajectories.
| CBA | CBA | CBA | ACB | Selection 3 |
<|MaskedSetence|> If the prediction probability exceeds 0.5, the sample is considered to belong to the positive class; otherwise, it is considered a negative class sample. <|MaskedSetence|> Threshold Adjustment is the method that is used to tune the prediction probability threshold to optimize the metric that is suit... | **A**: The Algorithm 1 shows the pseudocode of the Threshold Adjustment approach.
**B**:
2.3.1 Threshold Adjustment
In balanced binary data sets, the prediction probability threshold is normally set at 0.5.
**C**: However, for imbalanced datasets, setting the threshold at 0.5 can lead to undesirable results [46].
| BCA | BCA | CAB | BCA | Selection 4 |
<|MaskedSetence|> Given a circuit design HDL code DD, it can be mapped to a circuit graph representation GG through a bijection function f:D↔Gf:D\leftrightarrow G using our developed parser. <|MaskedSetence|> The node attributes include node type and width. <|MaskedSetence|> The width attribute reflects the output s... | **A**: DCG representation of HDL code.
**B**: Here, GG is a directed cyclic graph represented as (V,E,X)(V,E,X), where VV denotes the set of nodes, EE represents the set of edges with ei,j∈Ee_{i,j}\in E indicating a directed edge from viv_{i} to vjv_{j}, and XX represents the attributes of the nodes.
**C**: The node ... | ABC | CBA | ABC | ABC | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> By systematically evaluating a range of models, from traditional Support Vector Machines (SVMs) to deep learning architectures like LSTMs and CNNs, we identify the most effective approaches for this constrained environment. Our core contribution lies in the strategic application of... | **A**: Our experiments, conducted on a composite of the RAVDESS and SAVEE datasets, culminate in a model that not only achieves a new benchmark with 66.7%66.7\% accuracy and a 0.6310.631 F1 score but also provides a clear blueprint for developing powerful, data-efficient SER systems.
.
**B**: We propose and validate ... | CAB | CBA | CBA | CBA | Selection 2 |
The experimental setup consists of two main components: the training data for the emotion classifiers and the simulation parameters that define agent behavior and interaction dynamics.
Datasets. <|MaskedSetence|> <|MaskedSetence|> Its cultural and demographic homogeneity makes it a suitable test case for cross-popu... | **A**: The Karolinska Directed Emotional Faces (KDEF) dataset (Lundqvist et al.,, 1998) provides high-resolution frontal face images of 70 actors under controlled conditions, each performing the seven basic emotions.
**B**: The emotion classifiers were trained on three widely used facial expression datasets: JAFFE, CK... | ABC | BCA | BCA | BCA | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> The privacy-preserving nature of personhood credentials ensures that while accountability is maintained, the underlying personal data remains protected.
In summary, the synergy between authenticated delegation and personhood credentials fortifies AI agent inter... | **A**: This traceability not only curbs the risk of coordinated deceptive behavior but also provides service providers with a robust mechanism for enforcing rate limits or suspending access when an agent’s behavior violates defined policies.
**B**: Furthermore, embedding personhood credentials into authenticated deleg... | BCA | CBA | BCA | BCA | Selection 4 |
2 Related Work
Ensemble adversarial training primarily targets two key research directions: promoting diversity in model outputs and minimizing the transferability of adversarial examples across different sub-models. <|MaskedSetence|> <|MaskedSetence|> Transferability Reduced Smooth (TRS) [8] simultaneously promote... | **A**: For instance, Adaptive Diversity Promoting (ADP) [6] introduced a regularization method that encourages variability in non-maximum predictions.
**B**: Gradient Alignment Loss (GAL) [7] aimed to minimize the overlap of adversarial subspaces between sub-models.
**C**: LAFED [13] minimized the similarity between ... | ABC | ABC | ABC | ABC | Selection 3 |
<|MaskedSetence|> <|MaskedSetence|> A chronological split ensures that the temporal structure of the financial data is preserved for training and evaluation, thus avoiding data leakage.
At the heart of the methodology is a custom Gym-compatible environment designed to simulate financial allocation decisions. This env... | **A**: With each environment step, the system updates a Dirichlet belief vector—modeled as a prior over three budget categories—based on the agent’s actions, resulting in a Bayesian belief update.
**B**: III Methodology
The proposed methodology outlines a robust and modular framework for financial decision-making un... | BCA | BAC | BCA | BCA | Selection 3 |
The architecture of our proposed framework is illustrated in Figure 1. It is an end-to-end pipeline that begins with a natural language problem description as its primary input.
The initial step of this pipeline is Stage 1: LLM-driven Problem Structuring. <|MaskedSetence|> <|MaskedSetence|> The LLM then synthesize... | **A**: The process involves guiding the LLM to parse the input text and identify five categories: sets, over which indices are defined; parameters, the constants of the problem; decision variables, including their types (e.g., binary, continuous); the objective function with its optimization direction; and the constrai... | CAB | CAB | ACB | CAB | Selection 2 |
<|MaskedSetence|> Quantum computing’s ability to represent and entangle high-dimensional data offers a novel way to capture the complex interdependencies of musical features. Kashani et al. (2022) [4] implemented a note detection algorithm based on the Quantum Fourier Transform (QFT). Miranda et al. (2021) [5] envisio... | **A**:
While classical ML approaches for voice grading are promising, nascent technologies like quantum computing open new frontiers for audio analysis.
**B**: These features capture pitch and intonation accuracy, frequency stability (jitter) and amplitude stability (shimmer), LUFS energy (loudness and dynamics), and... | ACB | ACB | ACB | ABC | Selection 2 |
Transforming the dataset into a usable form required more than standardization – it demanded close reading of how the data had been entered and maintained over time. <|MaskedSetence|> By analyzing fields such as CHECK_BY annotations and other routine notes, it became possible to identify patterns of institutional lab... | **A**: These behaviors, rather than any predetermined structure, formed the foundation for parsing and reorganizing the dataset.
A schema was gradually developed to accommodate overlapping temporalities – planting, inspection, removal – and to represent multiple actors, including plants, curators, and institutional s... | CAB | CAB | BCA | CAB | Selection 1 |
Acoustically, harmony complements the melody by introducing a layer of complexity to music. It focuses on the simultaneous sounding of multiple notes, forming so-called chords; the notes of these chords are played in parallel (simultaneously) to the notes of the main melody. <|MaskedSetence|> <|MaskedSetence|> The s... | **A**: Continuing our mathematical framework, chords can be considered as combinations of note symbols.
**B**: In our "musical alphabet", a chord like “C-E-G” consists of symbols representing individual notes played together (in addition to the main melody).
**C**: Melodies are often played over a backdrop of harmoni... | CBA | ABC | ABC | ABC | Selection 3 |
Despite the limitations of metabolic models, their predictions have aligned well with experimental data from species that can be grown in laboratory conditions Harcombe et al. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> (2021); Libby et al. (2019); Souza et al. (2024).
The Friend or Foe data compendium can... | **A**: (2023b); Smith et al.
**B**: for more general ecology/evolution questions Libby et al.
**C**: (2014).
For other species, it is common practice to assume that their models may not be as accurate and to use them more conservatively, e.g.
| CBA | CBA | ACB | CBA | Selection 1 |
<|MaskedSetence|> Figure 2 illustrates two contrasting scenarios. In a difficult setting with sparse rewards (Figure 2(a)), a fixed exploration strategy (α>0\alpha>0) is superior. By lowering its certainty, the policy explores a wider action space and successfully discovers the distant reward mode. <|MaskedSetence|> ... | **A**:
Qualitative Analysis.
**B**: By increasing its certainty, the policy quickly hones in on the obvious optimal action.
.
**C**: Conversely, in an easier setting with dense rewards (Figure 2(b)), a fixed exploitation strategy (α<0\alpha<0) converges much faster.
| ACB | ABC | ACB | ACB | Selection 3 |
<|MaskedSetence|> Figure 7 shows the OTA throughput results for both CPU and GPU implementations across different configurations (CPU vs GPU with/without the --continuous-tx option). The x‐axis labels the configuration and the y‐axis indicates the achieved system throughput. <|MaskedSetence|> However, --continuous-tx... | **A**:
CPU- vs GPU-based iDFT/DFT.
**B**: In OAI –particularly when using a USRP X310 – enabling --continuous-tx is crucial to avoid power leakage from TX into RX and thus achieve high throughput.
**C**: As shown, the CPU version outperforms the GPU version in both downlink (DL) and uplink (UL) throughput, which al... | BCA | ABC | ABC | ABC | Selection 2 |
The total number of user input prompts used in this experiment is 20. <|MaskedSetence|> The prompt set covers various music styles and instrument types, providing standardized input support for system verification and comparative experiments. The complete prompt set can be found in the appendix. An example prompt is... | **A**: These prompts are all from the prompt set constructed by ComposerX, among which 10 have been specifically abridged, to leave more freedom for system creation.
**B**: The 8-bit synth should provide nostalgic, catchy melodies reminiscent of classic video games, while the electronic drums should add a rhythmic, up... | ACB | BCA | ACB | ACB | Selection 4 |
This paper presented a techno-economic framework for the evaluation of 5 wind-electrolyser use cases. <|MaskedSetence|> <|MaskedSetence|> Co-located configurations also outperform set-ups using the public grid as back-up, while the grid-only electrolyser remains the most costly option. The viability of curtailment-b... | **A**: Across all use cases, the electricity price and the electrolyser cost emerge as the primary LCOH drivers. Overall, the attractiveness of green hydrogen would improve with subsidies that aim to reduce the electricity cost, network charges and environmental levies, while policy-makers should facilitate wind-electr... | CBA | CBA | CAB | CBA | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> In addition, the two copies of tiles must have stripes pointing in different directions. Finally, the rows/columns that do not hold stripes must be numbered with the same number. <|MaskedSetence|> As our convention, we take the first copy to denote the horizontal direction and the... | **A**: Specifically, each copy forms a striped pattern where each stripe is numbered in a cyclic sequence.
**B**: Now a translation-invariant Hamiltonian can be simulated by using these tilings as a guideline for which sites are above, below, left, or right of each other.
**C**:
Figure 3: An example of how a 2D latt... | CAB | CAB | BAC | CAB | Selection 2 |
<|MaskedSetence|> We conduct several experiments where all parameters are set to their default values, and we apply one or more pruning techniques (for EA_T-GNN).
Table IV presents the experiment results. <|MaskedSetence|> In contrast, when all pruning methods are applied, the processing times reduces significantly (... | **A**:
VI-C5 Effectiveness of pruning techniques
In this section, we demonstrate the effectiveness of the pruning techniques in terms of number of POIs retrieved.
**B**: When two pruning methods are combined, the processing time improves as the algorithm explores fewer number of POIs.
**C**: We observe that without... | ACB | ABC | ACB | ACB | Selection 1 |
<|MaskedSetence|> Instead of maintaining a constant production (like the previous scenario), the electrolyzer adjusts its production rate based on the real-time life cycle carbon intensity of the grid’s incoming electricity. More specifically, and as shwon in Fig. <|MaskedSetence|> In effect, the rule defines discret... | **A**: 15, a predefined production rule is applied so that the carbon intensity (on the x-axis) is mapped to a production rate on the y-axis.
**B**: This production rule serves as a positive step toward environmentally responsive production and provides a comparative basis for understanding how time-varying production... | CAB | CAB | CAB | BAC | Selection 1 |
<|MaskedSetence|> Traditional object detectors, such as Faster R-CNN Ren (2015) and YOLO Redmon (2016), are effective for visual localization and classification tasks. Traditional models are confined to fixed labels and cannot handle open-ended, context-aware questions.
Vision-language models, by aligning images with ... | **A**: By systematically evaluating pre-trained VLLMs, Waste-Bench highlights their baseline capabilities and limitations, offering actionable insights to guide the improvement of future VLLMs..
**B**:
Despite advancements in Vision-Language Models (VLLMs), their application in complex, cluttered environments remain... | BCA | CAB | BCA | BCA | Selection 1 |
Additionally, the advent of VLMs [Li et al.(2022)Li, Li, Xiong, and Hoi, Sun et al.(2024)Sun, Wang, Yu, Cui, Zhang, Zhang, and Wang, Radford et al.(2021)Radford, Kim, Hallacy, Ramesh, Goh, Agarwal, Sastry, Askell, Mishkin, Clark, Krueger, and Sutskever] opened the possibilities for performing text-to-image retrieval in... | **A**: Additionally, previous works like [Tong et al.(2024b)Tong, Liu, Zhai, Ma, LeCun, and Xie, Shen et al.(2024)Shen, Xiong, Zhao, Wu, Chen, Zhu, Liu, Xiao, Varadarajan, Bordes, Liu, Xu, Kim, Soran, Krishnamoorthi, Elhoseiny, and Chandra, Tong et al.(2024a)Tong, Brown, Wu, Woo, Iyer, Akula, Yang, Yang, Middepogu, Wan... | CAB | CAB | CAB | CAB | Selection 2 |
To achieve a consistent subdivision of this space and optimize computational efficiency, we use a spherical icosahedron inscribed within S𝐩S_{\mathbf{p}}. Each face of the icosahedron serves as the base for a three-dimensional spatial bin that extends inward to the center 𝐩\mathbf{p}. <|MaskedSetence|> Each bin is ... | **A**: Furthermore, this implementation has achieved an impressive speed increase, more than 10 times faster than the manual spatial bin lookup in 3D space.
**B**: Our approach has demonstrated robust accuracy, with the lookup table yielding 99.6% precision in uniform points testing.
**C**: Given that an icosahedron ... | CBA | CBA | CBA | BAC | Selection 1 |
3.2 Hypothesis and claim extraction
Here, the goal is to automatically identify hypotheses from a scientific document. Because hypotheses can be viewed as claims prior to testing, the input, output, and methods for extracting hypotheses and claims are similar. <|MaskedSetence|> In White et al. <|MaskedSetence|> <|M... | **A**: The input document can be an abstract, e.g., [42, 43] or a full paper [28].
**B**: [44], authors propose and apply a schema for annotating sentences in full text of scientific articles into 9 types: hypothesis; goal; motivation; background; method; experiment; result; observation; and conclusion [44].
**C**: I... | ABC | ABC | ABC | ACB | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> In multi-hop question answering (HotpotQA), its knowledge-aware bandit-based routing dynamically selects relevant agents, enhancing multi-hop reasoning. For programmatic tasks (HumanEval), HiVA’s topological optimization ensures efficient agent collaboration for robust code generat... | **A**: These observations highlight HiVA’s ability to synergistically leverage its components across diverse tasks.
**B**: For mathematical reasoning (MATH), we observed that HiVA’s textual gradient mechanism effectively refines intermediate solutions, improving accuracy.
**C**: 4.3 Qualitative Case Study
To gain de... | CBA | CBA | CBA | CAB | Selection 3 |
The ”Needle in a Haystack” benchmark is widely used to evaluate the long-context retrieval capabilities of Large Language Models (LLMs). <|MaskedSetence|> The TConstFormer architecture proposed in this paper focuses its core contribution on demonstrating a fundamental improvement in computational and memory efficienc... | **A**: Extending the TConstFormer architecture to larger-scale models to explore its potential in complex retrieval tasks is a promising direction for future research.
.
**B**: However, this test primarily measures the complex instruction-following and long-range dependency abilities that emerge after pre-training o... | BCA | BCA | BCA | BCA | Selection 4 |
<|MaskedSetence|> The approximation result from the WoS-NN method is as Fig. <|MaskedSetence|> To validate our results, we have experimented with the Finite Difference Method on the same environment as Fig. 8(a). The error distribution of WoS-NN on this L-shaped region, referring to the FDM result, is given as Fig. ... | **A**: 8(c).
**B**:
Here, the Laplace equation has no known closed-form solution due to the irregular shape of the region.
**C**: 8(b).
| BCA | BCA | CAB | BCA | Selection 4 |
<|MaskedSetence|> 2025). Despite their powerful reasoning capabilities, these models often exhibit an “embodied reasoning gap” (Li et al. <|MaskedSetence|> Because they are trained on disembodied internet data, their generated plans can be disconnected from an agent’s physical capabilities and the constraints of a 3D... | **A**: 2024b).
**B**: Our work aims to bridge this gap by embedding geometry-aware priors directly into the model’s reasoning process.
.
**C**:
Recently, Large Language Models (LLMs) and Vision-Language Models (VLMs) have been leveraged for high-level planning in robotics, capitalizing on their vast common-sense k... | CAB | ACB | CAB | CAB | Selection 3 |
Table 1: Normalized throughput improvement over random walk (RW) across models and context-lengths on H100 (mean over 10 runs; higher is better).
Setups. We evaluate parallelization strategy performance using an in-house roofline-based simulator444Validated against Megatron-LM heuristic parallelization strategy.. ... | **A**: For all searches we allow a budget of 4000 agent forward + simulator calls (including invalid configurations), which takes less than 10 minutes.
**B**: This yields a search space of ∼109\sim\!10^{9} joint configurations.
**C**: The hardware system used in the evaluation is NVIDIA H100.
| CBA | CBA | CBA | ABC | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> For each dataset and fine-tuning configuration, Table 4 shows the highest detection rate achieved on L2. The MLP-Mixer model reached the best performance (132/138), having a recall of 95.65%, when trained with C21 and partial fine-tuning (𝒜2\mathcal{A}_{2}).
Inference-time compu... | **A**: Both the parameter count and FLOPs are computed using ptflops (Sovrasov, 2024) and are visualized in Figure 10.
**B**:
We evaluate trained models in inference mode to classify test subset L2, in order to quantify how many lenses we are able to recover from those identified in HSC PDR2 images of GAMA09H by More... | BCA | BAC | BCA | BCA | Selection 4 |
Human expert mapping to detailed secondary datasets addresses the granularity problem but creates new scalability and consistency challenges that limit practical implementation. A widely used activity-based data source – ecoinvent [19] – has upwards of 2,000 unique reference products. <|MaskedSetence|> Inconsistency a... | **A**: Time constraints make comprehensive mapping prohibitively expensive—with expert practitioners requiring 10+ minutes per material row, companies with large portfolios (often 10,000+ unique materials) [52] face high mapping costs.
**B**: This creates systematic bias in carbon footprint comparisons between product... | ABC | ABC | CAB | ABC | Selection 4 |
We now illustrate the framework by putting it to use in making some simple but clear claims about a collection of English-language album reviews from the online music publication, Pitchfork, and the probabilistic topic model, latent Dirichlet allocation (LDA) (Blei et al., 2003). While there is room for human interpret... | **A**: Prior to being text, each review may have been represented by the mechanics of keys being typed on, and in turn, those physical keyboard states can be understood as representations of the reviewer’s thoughts in response to an album.
**B**: For computational convenience, we let SS be a subset of 1,000 reviews fr... | CAB | CAB | CBA | CAB | Selection 4 |
The Transformer architecture [7] proposed a self-attention mechanism that captures pairwise interactions between all tokens in a sequence. <|MaskedSetence|> This property allowed Transformers to outperform in NLP and (subsequently) forecasting, but also introduced a quadratic cost in both time and memory, which is pr... | **A**: The Reformer [13] uses locality-sensitive hashes (LSH) to approximate attention, reducing the complexity from quadratic to almost linear and reversible residual layers to minimize memory usage during training.
**B**: Autoformer[8] introduces an auto-correlation mechanism instead of attention, which captures per... | CBA | CBA | CBA | BCA | Selection 2 |
In this paper, we introduce an implementation of monotone BART for binary outcomes. <|MaskedSetence|> <|MaskedSetence|> In Section 3 we propose probit monotone BART, including the model set up and some details on the code implementation. <|MaskedSetence|> Section 5 concludes.. | **A**: We first review the original, probit, and monotone BART models from the literature.
**B**: In Section 4 we demonstrate the proposed method with a simulation study.
**C**: This is done with a probit link, implemented using the ideas of normal latent variables/data augmentation from Albert and Chib, (1993), akin... | CAB | BAC | CAB | CAB | Selection 4 |
Table 1: Failure Rate % (↓\downarrow): Lower is better. AVGc = average over categories, AVGs = average over all samples. <|MaskedSetence|> We run each object for 30 steps in simulation, and 30 samples are given for the verifier to choose from at each step. <|MaskedSetence|> <|MaskedSetence|> | **A**: We see HAVE’s overall performance increase compared with baselines and ablation choices, as well as a more balanced ability across categories.
**B**: All 22 categories are treated as train categories, and we test on unseen instances.
**C**: Icon-to-text correspondence in Fig. 16.
.
| BAC | BAC | BAC | ACB | Selection 2 |
To address this challenge, we propose employing an SM to explicitly decompose the question-answering process into distinct subtasks based on the type of question. Specifically, relevant contextual information, such as the scene graph and question, is initially stored in the belief. Then, the LLM is prompted to determ... | **A**: If identified as a counting question, the LLM is subsequently prompted to extract relevant objects from the scene graph (extractObjects action upon entering the ObjectExtraction state).
**B**: Finally, the output of SHERPA is defined as the result of the final executed action.
.
**C**: Throughout this procedu... | ACB | ACB | BAC | ACB | Selection 2 |
For Echo, the embedding is derived from the last-layer hidden states before generating each token in the second occurrence of xx. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> Although Zhuang et al. (2024) also experimented with the average over multiple output tokens, we choose the first-token representati... | **A**: While Springer et al.
**B**: For PR, the embedding is taken from the last-layer hidden state before generating the first output token.
**C**: (2024) also considered using the last token representation, the mean token representation was found to be more effective in zero-shot scenarios, so we adopt it here.
| ACB | CAB | ACB | ACB | Selection 4 |
2. <|MaskedSetence|> (1996); Garcia-Molina et al. (2000); Console et al. <|MaskedSetence|> <|MaskedSetence|> To model this, researchers have proposed bag-extended relational algebras with well-defined operators for union, join, difference, and projection. To further capture ordering in query optimization, list-based... | **A**: Related Work
Relational Algebra for SQL.
Relational algebra serves as the formal backbone of structured query languages, providing a logical framework for reasoning about query correctness, equivalence, and transformation Grumbach et al.
**B**: Classical relational algebra assumes set semantics, but practical ... | ACB | ACB | ABC | ACB | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> Rather than considering arbitrary systems of curves in the plane, which may cross each other infinitely many times or even cover nonzero area, it is convenient to consider certain more well-behaved but still fully general string representations. A Jordan arc is a curve ambient isot... | **A**: We define a proper string representation to be a string representation in which each string is a Jordan arc, each two strings have finitely many points of intersection, at most two strings intersect at any point, and each intersection point is either a crossing of two strings or an endpoint of at least one of th... | BCA | BCA | BCA | ABC | Selection 1 |
Figure 7 and table 3 show the detailed performance analysis for three representative tensors. Because deli-3d is highly sparse with limited data reuse, its performance using ALTO is largely bounded by main memory bandwidth. <|MaskedSetence|> Although reddit is a low-density tensor, it has a large number of nonzero el... | **A**: Consequently, the performance of reddit using ALTO is bounded by L33 cache bandwidth.
**B**: Nevertheless, our ReLATE agent discovers a sparse encoding that significantly decreases L33 miss ratio, reducing main memory volume by 43%43\%.
**C**: Hence, its factorization data is largely served from L22 cache, as ... | BAC | ACB | BAC | BAC | Selection 3 |
<|MaskedSetence|> Visual overlays and metric evaluations demonstrated that VLM-enhanced refinement not only corrected structural drift but also improved local alignment in complex geometries, bridging the gap between automated contour generation and industrial-grade requirements.
Looking ahead, there remain several ... | **A**: First, expanding the dataset with additional industrial modalities (e.g., reflective alloys, composite materials) would further validate robustness across domains.
**B**:
The experimental analysis confirms that the proposed three-phase generative system can significantly reduce manual tracing effort, improve a... | BAC | ABC | BAC | BAC | Selection 1 |
We propose and develop an alternative approach that reduces the number of modules and still generates a smooth electric field profile with high resolution. <|MaskedSetence|> <|MaskedSetence|> Cascaded converters have previously used asymmetric voltage distributions [34, 35, 36, 37, 38].
Most asymmetric multilevel co... | **A**: Instead of equal voltage steps, our system assigns different operating voltages to each module.
**B**: Whereas in previous TMS technology the field granularity improved linearly with the number of modules, our approach achieves a near-exponential growth in resolution.
**C**: Some exceptions shrinked the differ... | BAC | ABC | ABC | ABC | Selection 2 |
<|MaskedSetence|> In conventional practice, organizations often deploy multiple disconnected tools to address these use cases, leading to fragmented workflows and inconsistent reporting. <|MaskedSetence|> <|MaskedSetence|> These improvements translate into direct economic benefits by lowering hardware and cloud reso... | **A**: SmartDiff eliminates this fragmentation by applying a consistent comparison framework across modalities, simplifying training and reducing the likelihood of errors in data validation.
Efficiency gains were also evident.
**B**: In controlled experiments, SmartDiff reduced memory usage by 40–50% relative to bas... | CAB | CAB | CBA | CAB | Selection 1 |
<|MaskedSetence|> In these approaches, formal characterizations are developed for hierarchies composed of a class of linear systems. The work presented in [23, 24] is based on approximate simulation and provides a formal treatment of a scenario that considers a bipedal robot and the canonical LIP model.
The hybrid z... | **A**: One tool for analyzing the connection between the FOM dynamics and their corresponding ROM abstractions is the approximate simulation approach [22] wherein a Lyapunov-like function, termed a simulation function, provides a quantitative metric characterizing the distance between trajectories of ROMs and FOMs.
**... | ABC | ABC | ABC | ABC | Selection 1 |
In order to approximate solutions to RTE, the upwind discontinuous Galerkin (DG) method is used. <|MaskedSetence|> A key strength of the high-order upwind DG method is its asymptotic preserving (AP) property, which ensures it correctly captures the diffusion limit of the RTE without resolving the small particle mean... | **A**: Besides upwind DG, DG methods have been actively developed for solving RTE.
**B**: DG has been a popular and powerful deterministic solver for RTE since its first introduced by Reed and Hill in simulating neutron transport [41].
**C**: These developments include hp-adaptive hybridized DG approach [15], approac... | BAC | BCA | BAC | BAC | Selection 3 |
Information Maximization has been widely used in machine learning and computer vision tasks including representation learning Tschannen et al. (2019); Hjelm et al. (2018); Bachman et al. (2019); Kemertas et al. (2020), deep clustering Hu et al. (2017); Jabi et al. (2019); Krause et al. <|MaskedSetence|> (2020a), doma... | **A**: (2023) and few-shot learning Boudiaf et al.
**B**: (2020), semi-supervised learning Chiaroni et al.
**C**: (2010), metric learning Boudiaf et al.
| CBA | ABC | CBA | CBA | Selection 1 |
The emergence of reasoning-capable models has introduced new challenges and opportunities in RLHF. <|MaskedSetence|> OpenAI’s o1-style reasoning models [24] pioneered this direction by incorporating sophisticated reasoning protocols into the RLHF framework. <|MaskedSetence|> Subsequently, models like SEED-1.5-Thinkin... | **A**: Following this breakthrough, DeepSeek-R1 [8] demonstrated the successful application of RLHF to reasoning models, showing significant improvements in mathematical and logical reasoning tasks.
**B**: Unlike traditional language models, reasoning models require careful balance between maintaining reasoning capabi... | BCA | CAB | CAB | CAB | Selection 2 |
To this end, we propose MorphGen, a
Morphology-Guided Generalization approach for robust domain generalization of Histopathological Cancer Classification, with a schematic overview illustrated in Figure 1. <|MaskedSetence|> Additionally, to further improve model robustness and convergence, we apply optimization techn... | **A**: Experimental results across three datasets—CAMELYON17, BCSS, and OCELOT—demonstrate that MorphGen learns robust and biologically meaningful representations, achieving consistently improved out-of-domain accuracy over strong baselines.
**B**: MorphGen incorporates histopathology images, their augmentations, and ... | BAC | BCA | BAC | BAC | Selection 3 |
<|MaskedSetence|> <|MaskedSetence|> Specifically, we draw inspiration from recently proposed continuous tempering approaches, in which Langevin dynamics is run over an augmented state space given by both position and a continuous temperature variable Gobbo & Leimkuhler (2015); Graham & Storkey (2017); Luo et al. (201... | **A**: Their success is based primarily on
the insight that such barriers are lowered when the temperature is raised (Marinari & Parisi, 1992; Swendsen & Wang, 1986; Hansmann, 1997; Sugita & Okamoto, 1999; Lenner & Mathias, 2016).
**B**:
To alleviate this problem, we turn toward tempering techniques, which are widel... | BAC | BAC | BAC | BCA | Selection 1 |
Despite these advances, the application of generative models in bioacoustics remains limited. <|MaskedSetence|> For example, Herbst et al. [12] compared denoising diffusion models and variational autoencoders (VAEs) for augmenting primate vocalizations, finding that traditional enhancement methods can sometimes match ... | **A**: Most prior work focuses on denoising or data augmentation rather than direct waveform synthesis.
**B**: However, high background noise in bird call recordings poses serious challenges for waveform-based models such as DiffWave [56] and WaveNet [57], which often fail under such conditions and produce audio lacki... | ACB | CAB | ACB | ACB | Selection 1 |
<|MaskedSetence|> <|MaskedSetence|> The conciseness gap in SciFact controls verbosity and grounding, while thematic overreach acts as a soft constraint to suppress conceptual drift in PrivacyQA. The explanation step during revisions consistently emerges as a key driver for identifying multiple reasoning pathways acro... | **A**: Dropping the coverage gap (in SciFact and PrivacyQA) notably reduces retrieval breadth but sometimes improves decision accuracy by encouraging more conservative outputs (PrivacyQA).
**B**: These results underscore that GIER’s gap-driven scaffolding not only improves explanation quality but also actively shapes ... | CAB | CAB | CAB | BAC | Selection 3 |
Fine-Tuning and Steerability. <|MaskedSetence|> <|MaskedSetence|> Investigating how steerable directions evolve with continued training is an ideal direction for future work.
Evaluation Scope and Generalization. <|MaskedSetence|> Extending to mobile and bimanual platforms, as well as unstructured environments, wi... | **A**: Our evaluations, while spanning both simulation and hardware, are limited to pick-and-place tasks with a robot arm.
**B**: We do not yet fully understand how VLA fine-tuning affects steerability.
**C**: While some semantic directions appear to retain causal influence after adaptation, it’s possible that fine-t... | ACB | BCA | BCA | BCA | Selection 3 |
<|MaskedSetence|> Accelerated Policy Convergence: The JEDP-RL framework achieved 3.2× faster convergence than PPO-RL (Fig. <|MaskedSetence|> <|MaskedSetence|> Enhanced Navigation Efficiency: Post-convergence analysis revealed JEDP-RL requires 25% fewer steps than PPO-RL to reach targets (Fig. 3(b)).. | **A**: 3(a)), validating the effectiveness of local Jacobian exploration.
**B**:
1).
**C**: This acceleration stems from the dual-phase architecture that decouples mechanical parameter identification from policy optimization, effectively addressing the time-varying kinematics problem in DCRs.
2).
| BAC | ABC | BAC | BAC | Selection 4 |
Realizing this goal presents three major technical challenges: (1) Homomorphic evaluation of state-of-the-art (SoTA) convolutional neural networks (CNNs) is computationally demanding due to the high multiplicative depth of CNNs [7, 9, 28, 1]. (2) Although several approaches [18, 28, 1] have demonstrated homomorphic ev... | **A**: A face image is divided into a grid of non-overlapping patches [14], and each is processed independently by a shallow patch CNN (PCNN).
**B**: Due to the lower resolution of individual patch, we reduce the multiplicative depth required for each PCNN.
**C**: Additionally, we design a distribution-aware low-degr... | ABC | ABC | ABC | BAC | Selection 3 |
<|MaskedSetence|> In their seminal work, Jordan, Kinderlehrer, and Otto observed that an implicit Euler discretization of the FPE can be reinterpreted as a variational problem: each timestep corresponds to minimizing a free energy functional that combines Shannon entropy with a Wasserstein-2 distance penalty [11]. <|... | **A**: Recall the forward evolution of the probability density ρ(x,t)\rho(x,t) under the FPE (2).
**B**: This insight, known as the JKO scheme, shows that the FPE can be understood as a gradient flow of entropy in the space of probability measures.
Building on this idea, Otto introduced a formal Riemannian calculus... | ABC | ABC | BAC | ABC | Selection 4 |
The construction cost of LEO satellite constellations is crucial for enabling efficient planning, budgeting, and the promotion of sustainable operations. <|MaskedSetence|> <|MaskedSetence|> Notice that the proposed cost model is an approximation rather than a generalized rule, the coefficients in the model can be rep... | **A**: For the manufacturing cost of a LEO satellite, it can be calculated as
ϖmanu=0.00185⋅Wsat,\varpi_{\text{manu}}=0.00185\cdot W_{\mathrm{sat}},.
**B**: In general, the costs associated with the space segment significantly contribute to the overall expenses of LEO satellite constellations.
**C**: Therefore, we... | BCA | BCA | ACB | BCA | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> We validate this from two perspectives: PC and mobile/edge device platforms.
PC Platform. In the NVIDIA GeForce RTX 4060 Ti platform, as shown in Tab. 6, even methods specifically designed for real-time fusion can achieve only quasi-real-time performance in limited scenarios. Thi... | **A**: In practical applications, real-time performance is crucial for algorithm usability.
**B**: The primary advantage of LUT-Fuse lies in its exceptional computational speed combined with competitive fusion quality.
**C**: Our proposed LUT-Fuse stands as the only method achieving real-time, and even super-real-tim... | ABC | ABC | ABC | ABC | Selection 4 |
We conducted an ablation study on six variants of our model across three unseen tasks: (1) ‘LLMDPD’, our full model, which includes all components; (2) ‘LLMDPD-OLMo-1B’, which replaces the base LLM with OLMo-1B (Groeneveld et al., 2024) for processing text prompt embeddings; (3) ‘w/o-prompt’, which removes both text a... | **A**: This study systematically evaluates each component’s contribution to overall performance.
The results of the ablation study are presented in Table 3.
**B**: However, this variant still maintains strong generalization performance.
**C**: The ‘w/o-prompt’ variant, which removes only the text prompt, results in... | ABC | ABC | ABC | ABC | Selection 1 |
<|MaskedSetence|> Our proposed method, STAR, consistently outperforms prior approaches across all target domains, achieving the highest average accuracy of 60.3%, surpassing the next-best method PR-C by a notable margin of 3.2%. STAR sets a new state of the art in the "Art" domain with 75.3%, outperforming ACVC; achie... | **A**: STAR achieves the highest average accuracy of 30.0%, outperforming the strongest prior method, AdvST, across all five target domains.
**B**:
Table 1 presents the classification results on the PACS dataset using ResNet-18 as the backbone.
**C**: The most pronounced gain is observed in the "Sketch" domain, wher... | BAC | BAC | CBA | BAC | Selection 2 |
<|MaskedSetence|> MR is often described by experts as a sliding scale between an entirely physical environment with no virtual elements and a fully virtual environment [8]. MR is not confined to the virtual environments of VR or the single mode of interaction found in AR. <|MaskedSetence|> Second, the digital content... | **A**: (b) The kinematic chain of the calibration system.
**B**: Parveaua and Addaa categorized MR into three types: First, it consists of both real and virtual content, allowing data contextualization.
**C**:
MR combines elements of both VR and AR, containing both physical and virtual components [21].
| CBA | CBA | CBA | CBA | Selection 1 |
II-B Multimodal Environmental Monitoring
Recent research has increasingly focused on multimodal fusion strategies for enhanced environmental monitoring. Gowthami et al. [13] proposed integrating satellite imagery with deep learning for Delhi’s AQI forecasting, achieving 14% improvement over single-modality approaches... | **A**: Sarkar et al. [15] demonstrated the viability of mobile-captured images for pollution alert systems, while Hameed et al. [16] proposed a deep multimodal architecture that fuses CCTV traffic imagery with sensor data for AQI estimation in Dalat City, Vietnam.
**B**: Our proposed framework addresses these gaps thr... | ACB | ACB | ACB | ACB | Selection 1 |
<|MaskedSetence|> Following the experimental setup of LRMR [5], the sub-image of size 200×400×166200\times 400\times 166 was used for testing.
Fig. 7 provides a visual comparison of the denoising results obtained by different methods on real-world noisy HSI. <|MaskedSetence|> <|MaskedSetence|> Although T3SC, TRQ3D... | **A**:
The first real-world noisy HSI was captured by the EO-1 satellite, with an original size of 400 ×1000×242\times 1000\times 242 and a spectral range from 400 to 2500 nm.
**B**: As shown in Fig. 7LABEL:sub@fig:eo1_clean_visual, the original HSI was severely degraded by complex noise such as strip noise and deadl... | ABC | ABC | ABC | BAC | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> We implement multi-scale temporal masking with varying durations of 2, 4, 8, and 16 frames to capture fine-grained temporal patterns. For surgical video context alignment, we adopt AdamW optimizer with the learning rate of 1×10−51\times 10^{-5} and the weight decay of 0.020.02. We ... | **A**: We randomly generate instrument-centric tube masks using the bounding boxes of surgical instruments, and set the probability rr as 10% to randomly keep a small proportion of video tubes as the hint for reconstructing the masked video contents.
We initialize the visual encoder with the weights of VideoMAE [37].
W... | CAB | CAB | CAB | ABC | Selection 1 |
Recent advances in robotic manipulation leverage large-scale, vision-language-action models [5, 6, 7, 8, 9]. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> Recent efforts to eliminate action labels through dense correspondence [19], goal-conditioned exploration [20], or stereo-based pose estimation [21] hav... | **A**: However, scaling such models is challenging due to the cost and effort required for human-annotated demonstrations.
**B**: Some approaches learn general representations for policy learning [10, 11] or label action-free dataset with latent action [12], while others guide actions by predicting intermediate visual... | ACB | ACB | ACB | BCA | Selection 1 |
Other researchers investigated matrix structures and their impact on signal propagation. For instance, Saxe et al. [13] advocated orthogonal initialization, leveraging the property 𝑾𝑾T=𝑰\bm{W}\bm{W}^{T}=\bm{I} to preserve signal norms through forward and backward passes in linear regimes. Orthogonal initialization... | **A**: Their analysis demonstrates that orthogonal weights enable depth-independent convergence, in contrast to Gaussian initialization, which requires network width to increase linearly with depth for efficient training.
**B**: Building upon this intuition, Hu et al. [22] provided a theoretical justification for orth... | BAC | BAC | BAC | CAB | Selection 2 |
The graph-based ANNS algorithms commonly use the greedy search algorithm for searching nearest neighbors. <|MaskedSetence|> Starting from the specified seed node, each candidate’s neighbors are examined. <|MaskedSetence|> The iterative distance calculation calls find nodes closer to the query, progressively moving t... | **A**: Neighbors whose distances to the query exceed the farthest distance in the candidate set are discarded (i.e., negative nodes), while others are included in the candidate set and further refined (i.e., positive nodes).
**B**: Illustration of the greedy search algorithm..
**C**: As shown in Figure 1, it maintain... | BCA | CAB | CAB | CAB | Selection 3 |
We further investigate different text embedding models for retrieval, using doubao-embedding-text Seed et al. <|MaskedSetence|> <|MaskedSetence|> <|MaskedSetence|> Table 8 examines the impact of different text embedding models on retrieval performance. Here, doubao-embedding-text Seed et al. (2025) emerges as the t... | **A**: (2025), gte-multilingual-base Zhang et al.
**B**: (2024a).
**C**: (2024), and multilingual-e5-large-instruct Wang et al.
| ACB | ACB | ACB | ACB | Selection 4 |
<|MaskedSetence|> <|MaskedSetence|> VPFC operates by recalibrating the confidence assigned to visual evidence during the mapping from visual features to semantic concepts, specifically with respect to object existence. This strategy effectively reduces omission hallucinations while avoiding the introduction of fabric... | **A**: In summary, our contributions are as follows:.
**B**:
In Section 3.3, we examine the mapping from visual features to semantic concepts through attention intervention experiments, investigating how the model constructs visual evidence to infer the presence or absence of objects.
**C**: Building on this analysi... | BCA | BCA | ACB | BCA | Selection 2 |
Evaluation Metrics. <|MaskedSetence|> <|MaskedSetence|> This metric is computed using the classifier from HarmBench (Mazeika et al., 2024). For visual understanding evaluation, we employ task-specific utility metrics. MM-Vet uses GPT-4 with few-shot prompts to generate utility scores ranging from 0 to 1, while SQA c... | **A**: For jailbreak detection, we quantify robustness using the attack success rate (ASR), defined as the proportion of successful jailbreaks among total attack attempts.
**B**: This study compare SPO-VLM against two baseline methods.
**C**: For toxicity assessment, we employ the Detoxify classifier (Hanu and Unitar... | CAB | CAB | CAB | CBA | Selection 2 |
<|MaskedSetence|> <|MaskedSetence|> For instance, R1-Searcher (Song et al., 2025b) and Search-R1 (Jin et al., 2025) both employ RL frameworks to teach LLMs how to interleave search queries with their reasoning steps, with the latter introducing techniques like retrieved token masking for more stable training. <|Mask... | **A**: Reinforcement learning (RL) is a common paradigm in this domain.
**B**: Training Agents for Search and Retrieval
Another prominent research direction focuses on explicitly training agents to interact with external information sources, most notably search engines.
**C**: AutoRefine (Shi et al., 2025b) presents... | BAC | BAC | BCA | BAC | Selection 4 |
<|MaskedSetence|> “Original” refers to the result without any advanced data augmentation. <|MaskedSetence|> In particular, we see improvements of 2.27% on CUB and 2.98% on Flower over CutMix. This is because our method mixes the estimated noise of two classes at each denoising step of Stable Diffusion, resulting in m... | **A**:
5.1 Quantitative Evaluation
Table 1 shows the classification results on the three datasets.
**B**: We observe that our proposed data augmentation method outperforms conventional methods that combine multiple images (CutMix, MixUp) on CUB and Flower.
**C**: As NoiseCutMix depends on SD, no improvements were o... | BAC | ABC | ABC | ABC | Selection 2 |
<|MaskedSetence|> (See Figure 1 for a diagram showing the relationships between the optimal solution sizes of the studied problems.) A feedback edge set of a graph GG is a set of edges whose removal turns GG into a forest. The smallest size of such a set, denoted by c(G)c\left(G\right), is the cyclomatic number of GG... | **A**: To find an optimal feedback edge set of a connected graph, it suffices to consider a spanning tree; the edges not belonging to the spanning tree form a minimum-size feedback edge set.
.
**B**: In this paper, we assume all our graphs to be connected.
**C**: Our goal.
Our objective is to study the three above ... | CBA | BAC | CBA | CBA | Selection 4 |
Unifying perturbation Strategies. <|MaskedSetence|> This inspires us to explore the specific connection between different perturbation strategies. As a starting point, we demonstrate that PerturbEdge, PerturbNode, PerturbWeight, and PerturbEmbedding can all be formulated in Table 1. <|MaskedSetence|> <|MaskedSetence... | **A**: More importantly, we find that PerturbEdge, PerturbNode, and PerturbWeight are actually special cases of PerturbEmbedding, and thus can be expressed in a uniform framework.
**B**: As shown in Figure 1, in intuition, applying perturbation to features, edges or weights will all ultimately act on the embedding 𝐇\... | BAC | ABC | BAC | BAC | Selection 3 |
Most existing token eviction strategies Zhang et al. <|MaskedSetence|> <|MaskedSetence|> (2024); Guo et al. (2024) reduce KV cache size via identifying important tokens based on their importance scores (e.g., attention scores).
Then, as shown in Figure 1, a static selection is utilized to remove the KV cache of the r... | **A**: (2023); Li et al.
**B**: (2024); Cai et al.
**C**: This observation highlights that
retaining multiple tokens with high importance can lead to redundancy if they are highly similar, while the other less important but dissimilar tokens that contain more diverse semantic information can contribute uniquely to th... | ABC | ABC | CBA | ABC | Selection 4 |
<|MaskedSetence|> In vanilla GCG, these steps are fully deterministic: tokens are chosen strictly by gradient magnitude and the final suffix is the one minimizing loss. <|MaskedSetence|> By contrast, our approach replaces both stages with temperature-weighted sampling. <|MaskedSetence|> This stochastic mechanism per... | **A**: While efficient, this greedy procedure often traps the optimization in local minima and reduces robustness across diverse prompts.
**B**:
Our method introduces temperature-based sampling at both the token-selection and suffix-update stages, aiming to balance greedy exploitation of gradient signals with probabi... | BAC | BAC | BAC | CBA | Selection 2 |
Addressing the gap between training data and real-world applications will likely require a combination of domain adaptation and transfer learning techniques [242, 243]. <|MaskedSetence|> Robust optimization methods, like adversarial training, can enhance model stability [244, 245]. A particularly promising direction ... | **A**: Neural field models offer efficient representation and rendering of dynamic acoustic scenes [94, 183].
.
**B**: Well-designed data augmentation strategies help simulate a wider range of real-world conditions during training.
**C**: Research on continual learning and online adaptation could enable models to a... | BCA | BCA | BCA | BCA | Selection 1 |
4 Experiments
Dataset.
We evaluated our method using the THuman4.0 dataset [90], a multi-view dataset with a resolution of 1330×11501330\times 1150, featuring characters with rich textures and dynamic details. <|MaskedSetence|> <|MaskedSetence|> Each sequence includes manually selected video clips featuring turning... | **A**: Foreground masks were obtained using Segment Anything 2 [55]..
**B**: To assess our method on higher-resolution images, we collected an additional dataset, named Mono2K, comprising images at the resolution of 1500×20481500\times 2048.
**C**: For evaluation, we selected all three sequences (subject00, subject01... | BCA | BCA | BCA | ACB | Selection 2 |
The core functionality of our knowledge assistant is to provide evidence-based answers from medical studies to posed medical questions. The system operates through a multi-stage pipeline, ensuring both broad retrieval and precise information extraction and synthesis. <|MaskedSetence|> <|MaskedSetence|> We use SciSpa... | **A**: After the user inputs a question, the first step is transforming it into a query optimized for PubMed’s search engine.
**B**: This is visualized in Figure 1.
Document Retrieval.
**C**: This is used to expand the search query with multiple options..
| CBA | BAC | BAC | BAC | Selection 3 |
Our contributions are as follows. We formulate the finite state finite action partially observable restless multi-armed bandit (PO-RMAB) problem. Our work is the first to study multi-action PO-RMAB.
We propose a Lagrangian relaxation technique using Lagrangian multipliers method for budget constraints. We describe the ... | **A**: We present the preliminaries and model description in Section II.
**B**: We develop two timescale stochastic approximation based approach for the Lagrangian bound computation.
**C**: We next present a study on Monte-Carlo rollout policy for Lagrangian bound in V, heuristic policies in VI, indexability and Whit... | BAC | CBA | BAC | BAC | Selection 4 |
Token reduction has been widely explored in both computer vision Meng et al. (2022); Chen et al. (2023); Pan et al. (2022); Ryoo et al. (2021); Rao et al. <|MaskedSetence|> <|MaskedSetence|> (2021). However, these methods usually require training, while our method can be done in a training-free manner. In multimodal ... | **A**: (2021) and natural language processing (NLP) Goyal et al.
**B**: (2020); Kim and Cho (2020); Lassance et al.
**C**: VisionZip Yang et al.
| ABC | ABC | ABC | BCA | Selection 1 |
<|MaskedSetence|> Kim et al. <|MaskedSetence|> <|MaskedSetence|> (2024); Kajitsuka and Sato (2024). More recently, Kajitsuka and Sato (2025) study the optimal number of parameters required for memorization.
. | **A**: (2023) prove that transformers with 2n2n self-attention layers
suffice for the memorization of length-nn inputs.
**B**:
Memorization Capability of Transformer.
There are many works on the memorization capabilities of the transformer architecture.
**C**: This result was improved to single-layer transformers i... | BAC | CAB | BAC | BAC | Selection 1 |
<|MaskedSetence|> As shown in Table 5, using only experts (FID:26.55) or gating (FID:31.30) yields partial improvements over the baseline (FID:33.25), while their combined use achieves optimal FID (22.24). This synergy arises because experts decompose facial semantics into multiple binary components (mask error drops ... | **A**:
We ablate the individual and joint effects of global-local experts and dynamic gating.
**B**: This validates that global-local experts and adaptive gating are mutually essential for high-fidelity and controllable generation.
.
**C**: Although the baseline shows marginally higher text alignment (Text:26.71 v... | ACB | BAC | ACB | ACB | Selection 1 |
Table 2 and Fig. 14 compare the tracking accuracy and the mapping quality of the baseline implementation SplatAM and the proposed AGS algorithm, respectively. We assess these aspects using metrics including ATE RMSE (Absolute Trajectory Error Root Mean Square Error), commonly employed to measure the accuracy of trackin... | **A**: Thus, by synergizing 3DGS with traditional SLAM techniques, AGS effectively balances enhanced performance with minimal accuracy trade-offs.
.
**B**: For mapping, the results show that AGS has an average of 2.36%2.36\% PSNR loss compared to SplatAM, whereas traditional SLAM methods lack the capability for photor... | ABC | CBA | CBA | CBA | Selection 2 |
<|MaskedSetence|> Our approach to the MIDI-B challenge highlights both the strengths and limitations of relying on transformer-based models for PHI detection. While AI-based methods such as RoBERTa demonstrate strong performance in identifying PHI in free-text contexts, their effectiveness diminishes when applied to s... | **A**: The development of a robust de-identification framework for DICOM files presents unique challenges, particularly in balancing patient privacy with data utility.
**B**: The iterative refinements applied throughout the challenge led to progressively improved results, underscoring the importance of a hybrid, adapt... | ABC | ABC | ABC | CAB | Selection 3 |
2.1. LLMs as Agents
Our work builds on recent advances in enabling large language models to act as agents that reason, plan, and take actions in interactive environments. <|MaskedSetence|> <|MaskedSetence|> Although MMSearch shares topical similarity with our work by requiring retrieval from multimodal sources, its... | **A**: Benchmarks such as AgentBench (Liu et al., 2024) evaluate agent performance across web navigation, tool use, and knowledge-intensive tasks, while MMSearch (Jiang et al., 2025) focuses on multimodal retrieval and summarization.
**B**: Chain-of-Thought (CoT) (Wei et al., 2022) promotes step-by-step reasoning, ReA... | BAC | BAC | BAC | BAC | Selection 1 |
End of preview. Expand in Data Studio
README.md exists but content is empty.
- Downloads last month
- 13