table_id stringlengths 16 17 | paper_id stringlengths 10 10 | paper_link stringlengths 32 32 | caption stringlengths 0 1.43k | text_mentions stringlengths 220 3.14k |
|---|---|---|---|---|
2305.07570_TAB_1 | 2305.07570 | https://arxiv.org/abs/2305.07570 | Experimental results on the Bottle Shampoo (604,903 input points). | In Tables~\ref{tab:ExperimentalResultsShampoo} to~\ref{tab:ExperimentalResultsToyBear}, we report both the results of the comparison algorithms and the result after these have been remeshed, indicated by~``(Re)". A first thing to notice when regarding the results presented in Tables~\ref{tab:ExperimentalResultsShampoo}... |
2305.07570_TAB_2 | 2305.07570 | https://arxiv.org/abs/2305.07570 | Experimental results on the Cloth Duck (1,018,891 input points). | Notable exceptions are the Bowl Chinese model, Table~\ref{tab:ExperimentalResultsBowl}, and the Cloth Duck model, Table~\ref{tab:ExperimentalResultsClothDuck}. However, the largest average edge length across all models is~$0.2181$ for our algorithm, attained on the Cloth Duck model, Table~\ref{tab:ExperimentalResultsCl... |
2305.07570_TAB_3 | 2305.07570 | https://arxiv.org/abs/2305.07570 | Experimental results on the Toy Bear (607,501 input points). | In Tables~\ref{tab:ExperimentalResultsShampoo} to~\ref{tab:ExperimentalResultsToyBear}, we report both the results of the comparison algorithms and the result after these have been remeshed, indicated by~``(Re)". A first thing to notice when regarding the results presented in Tables~\ref{tab:ExperimentalResultsShampoo}... |
2306.10843_TAB_1 | 2306.10843 | https://arxiv.org/abs/2306.10843 | Accuracy on the different test bottles used in the experiment. The column ``Day" indicates the day since the pupae were sexed. The case of 100\% females is not shown because its result is trivial. | The dataset used for this study consists of 2 hours and 8 minutes of recording. The recording of each container corresponds to a total of 32 minutes. Thus, there are 8 minutes of recording per container per day. It should be noted that the training sessions are independent among days as the mosquitoes may vary in their... |
2310.16283_TAB_1 | 2310.16283 | https://arxiv.org/abs/2310.16283 | Basic Descriptives per Variable (In Percent) | The box plots in Fig. \ref{fig1} illustrate some of the basic descriptive statistics of each variable in the COVID-19 era. These descriptives are based on the monthly rate of change in percentages. Basic descriptive statistics, in percents, are provided in Table \ref{tab1}. The mean monthly rate of change is reasonably... |
2311.00968_TAB_4 | 2311.00968 | https://arxiv.org/abs/2311.00968 | The $Hits@k$ scores and affective matching loss of the proposed method (AMT) and baseline models on the test set. | Table~\ref{tab:chordpred} shows the \( Hits@k \) scores and affective matching loss of our proposed Affective Multimodal Transformer (AMT) with and without affective matching loss, as well as the results for two baseline models: Transformer~\citep{vaswani2017attention} and Music Transformer~\citep{huang2018music}. Our ... |
2311.00968_TAB_5 | 2311.00968 | https://arxiv.org/abs/2311.00968 | Hyperparameter tuning experiment, reported on the validation set. The last three columns show the results for the listed parameters. | To tune the hyperparameters of our proposed Affective Multimodal Transformer (AMT) model, we conducted an experiment. Table~\ref{tab:ablexp}, summarizes this experiment, whereby we systematically vary the transformer parameters, including heads and layers. Through this experimental analysis, we identified the best-perf... |
2312.10045_TAB_1 | 2312.10045 | https://arxiv.org/abs/2312.10045 | An example showing how RCKT uses response influence approximation to conduct counterfactual reasoning. ``$\circ$" denotes the mask operation. | Tab.~\ref{tab:run} illustrates the process of RCKT inferring the same example after response influence approximation. As shown, $q_6$ is assumed to be answered correctly and is then flipped to an incorrect response. The probabilities of correctly answering the questions with correct responses ($q_1, q_3$, and $q_4$) be... |
2405.20397_TAB_1 | 2405.20397 | https://arxiv.org/abs/2405.20397 | Input features with corresponding notations, references, and descriptions | The features are carefully chosen from previous physics-based modeling research regarding adsorption energy computation. The features are categorized into two groups: adsorbate-specific and catalytic-specific features, as seen in Table \ref{tab:shap_values}. Each feature provides insights into electronic structure, bon... |
2405.20434_TAB_2 | 2405.20434 | https://arxiv.org/abs/2405.20434 | Users’ ratings of (a) their trust in the model and (b) the ease of assessing the accuracy of the model's response, for the baseline and each of the factuality score designs. \textit{t} and \textit{p} values were calculated within the omnibus statistical model, and represent the comparison of each Design Strategy agains... | First, we compared users’ ratings of their trust of the model that produced the response, for each of the designs compared against the baseline. All six of the designs were rated as significantly more trustworthy compared to the baseline, suggesting that all of the designs that presented response factuality increased u... |
2405.20590_TAB_1 | 2405.20590 | https://arxiv.org/abs/2405.20590 | Sample sizes and imbalance ratios in each partition in SWAN-SF. The time spans between two adjacent partitions are not overlapping. | The SWAN-SF dataset is created through a sliding-window methodology with a 1-hour step size, leading to temporal coherence issues that can introduce bias and overfitting during model training \cite{ahmadzadeh2021train}. To address this, SWAN-SF is divided into five partitions without temporal overlapping with each part... |
2405.20590_TAB_2 | 2405.20590 | https://arxiv.org/abs/2405.20590 | DtP of different data augmentation methods on each testing partition. Each cell shows the mean and standard deviation (in the parenthesis) across ten runs of our random undersampling experiments. | The DtP metric ranges from 0 to $2\sqrt{2}$, where 0 denotes a perfect model and $2\sqrt{2}$ signifies a classifier that assigns the opposite labels to all testing instances. Tab.\ref{tab:exp_res} shows the mean and standard deviation (in the parentheses) of DtP for different augmentation methods across ten runs of ran... |
2405.20643_TAB_2 | 2405.20643 | https://arxiv.org/abs/2405.20643 | Error (º) in the ETH-XGaze test set when trained with limited data and different kinds of augmentations and DNN architectures~\cite{zhang2020eth,abdelrahman2022l2cs,gazetr2022}. | Table~\ref{tab:exp1_error} shows the error obtained by the DNN in the test set when trained with the different augmented sample configurations, as well as when trained without augmentations. Fig.~\ref{fig:aug_bothdoms} shows qualitative results of the different augmentations, as evaluated in Table \ref{tab:exp1_error}... |
2405.20643_TAB_3 | 2405.20643 | https://arxiv.org/abs/2405.20643 | Gaze preservation between domains when different components are frozen, Gaze-aware Local Generator ($GLG$) or Render Net ($R$), and whether mask constraint is applied. | In Table \ref{tab:ablation_fr}, we observe that freezing only the shape-oriented layers in the GLG is not enough, the weights of the whole GLG need to be preserved. The Render Net $R$ also contains some information related to the gaze, as it is required to freeze the 2 initial residual blocks to transfer the gaze. It i... |
2405.20682_TAB_1 | 2405.20682 | https://arxiv.org/abs/2405.20682 | Optimisation results summary - CS 1-1 CIGRE LV network | Table \ref{tab:hc_cs_1_1_cigre} shows the summarised optimisation results of the static export limits calculation in the modelled CIGRE LV network. As can be seen, the computational time values are similar in all scenarios and less than 0.3 seconds in all scenarios, with the exception of Scenario 1, in which the optimi... |
2405.20682_TAB_2 | 2405.20682 | https://arxiv.org/abs/2405.20682 | Optimisation results summary - CS 1-2 CIGRE LV network | The summarised results of the HC calculation in the case of static import limits in the modelled CIGRE LV network are presented in Table \ref{tab:hc_cs_1_2_cigre}. All the conclusions drawn from the analysis of the results when calculating the static export limits remain the same in this case. The computational time va... |
2405.20682_TAB_3 | 2405.20682 | https://arxiv.org/abs/2405.20682 | Optimisation results summary - CS 2-1 CIGRE LV network | Table \ref{tab:hc_cs_2_1_cigre_lv} presents summarised results of calculating dynamic export limits in the CIGRE LV network. In this case, there are no scalability issues, even in S1, in the case of MINLP formulation, which takes more than 10 minutes to solve the defined optimisation problem. However, given the definit... |
2405.20682_TAB_4 | 2405.20682 | https://arxiv.org/abs/2405.20682 | Optimisation results summary - CS 2-1 Real-world LV feeder | The complexity of the optimisation problem increases with the size of a modelled network and the number of time intervals, as seen from the results in Table \ref{tab:hc_cs_2_1_rw_feeder} and Table \ref{tab:hc_cs_2_2_rw_feeder}, presenting the summarised optimisation results in the case of calculating dynamic export and... |
2405.20682_TAB_5 | 2405.20682 | https://arxiv.org/abs/2405.20682 | Optimisation results summary - CS 2-2 Real-world LV feeder | The complexity of the optimisation problem increases with the size of a modelled network and the number of time intervals, as seen from the results in Table \ref{tab:hc_cs_2_1_rw_feeder} and Table \ref{tab:hc_cs_2_2_rw_feeder}, presenting the summarised optimisation results in the case of calculating dynamic export and... |
2405.20700_TAB_2 | 2405.20700 | https://arxiv.org/abs/2405.20700 | Average diagnosis accuracy for different methods under four settings. | We average the diagnosis accuracy of two domain adaptation cases and list them in Table \ref{tab:CWRUacc}. The classic DA approaches JDA and DANN reach good accuracy for the B2B setting but their accuracy drops dramatically for the three imbalanced settings because they do not consider the imbalance. With the help of c... |
2405.20700_TAB_3 | 2405.20700 | https://arxiv.org/abs/2405.20700 | Dataset information of three-phase flow process | Cranfield University's three-phase flow facility is designed to deliver precise, controlled flow rates of water, oil, and air into a pressurized system \cite{ruiz2015statistical}. The facility can manage either single phases of these fluids or their combinations in specific proportions. During this experimentation, two... |
2405.20700_TAB_4 | 2405.20700 | https://arxiv.org/abs/2405.20700 | Average diagnosis accuracy for different methods under four settings | Table \ref{tab:Tripleacc} lists the average diagnosis accuracy of two domain adaptation cases, i.e., A to B and B to A. We can see that the proposed method achieves the best performance in general, especially for the B2I and I2I settings. One can see that in the I2I setting, the classific DA approaches almost fail beca... |
2405.20700_TAB_5 | 2405.20700 | https://arxiv.org/abs/2405.20700 | Ablation study for the proposed method | We also implement ablation studies to test the efficiency of IaCLR and BaADA separately based on the 1797 $\xrightarrow{}$ 1730 setting. The benchmark is chosen as the original DANN. We add two key parts of the Sd-CDA step by step on the basis of DANN and generate three situations, i.e., DANN + Ia-CLR, DANN + Ba-ADA, a... |
2405.20731_TAB_1 | 2405.20731 | https://arxiv.org/abs/2405.20731 | Performance of the three models at different resolutions, in degrees Celsius. | The results, displayed in Table \ref{tab:performance}, represent the average MAE for the year 2023. It is clear from the results that the linear regression model, despite its simplicity and reduced computational complexity, does not appear to be adequate for this type of prediction. The MAE values for the linear regres... |
2405.20830_TAB_1 | 2405.20830 | https://arxiv.org/abs/2405.20830 | Open LLM Leaderboard Evaluation. | In our comprehensive analysis of the SAPO framework, detailed in Tables~\ref{tab:openlead} and~\ref{tab:ifeval}, we systematically evaluate and compare the performance of various models, focusing on multiple dimensions crucial for the performance of LLMs. Table~\ref{tab:openlead} presents a detailed comparative perform... |
2405.20830_TAB_2 | 2405.20830 | https://arxiv.org/abs/2405.20830 | Evaluation on IFEval, MT-Bench, AlpacaEval 2.0. | In our comprehensive analysis of the SAPO framework, detailed in Tables~\ref{tab:openlead} and~\ref{tab:ifeval}, we systematically evaluate and compare the performance of various models, focusing on multiple dimensions crucial for the performance of LLMs. Table~\ref{tab:ifeval} evaluates language model alignment perfor... |
2405.20877_TAB_1 | 2405.20877 | https://arxiv.org/abs/2405.20877 | Waveform Approximation | To facilitate the testing of our results without the need to create the DNN architecture, we provide the curve fitting parameters for the DNN-based waveforms of \ref{fig:oddRollOff} in Table \ref{table:TableCoef}. This may also allow practical deployment in real communication systems, since curve fitting can generate w... |
2405.20877_TAB_2 | 2405.20877 | https://arxiv.org/abs/2405.20877 | MSE Improvement Table | % In Table \ref{table:TableMSE} the percentage improvement gain of the DNN-generated waveforms over their RC and BTRC counterparts is presented for different roll-off factors and sampling error. As observed the proposed framework can generate waveforms that provide considerable gain, but especially for mid-range roll-o... |
2405.20982_TAB_2 | 2405.20982 | https://arxiv.org/abs/2405.20982 | Different model slicing methods implemented for evaluation. | We built a prototype of \name in C++ and Java, as well as three other data plane verification methods representing the prior techniques, as shown in Table~\ref{tab:taxonomy}. All verification methods are implemented with Apache Samza~\cite{2017-NoghabiEtAl} for task distribution, FoundationDB~\cite{2021-ZhouEtAl} for i... |
2405.20982_TAB_3 | 2405.20982 | https://arxiv.org/abs/2405.20982 | Network datasets. | Our network dataset consists of 5 networks that are used in different enterprise and commercial settings, listed in Table~\ref{tab:dataset}. N1 is a small example network containing virtualized devices and designed for system tests of a commercial network management suite~\cite{vmware-aria}. N2 is a lab network with re... |
2405.20982_TAB_4 | 2405.20982 | https://arxiv.org/abs/2405.20982 | Number of rules modeled by the loaded slice. | To quantify the complexity of the slices, we calculate the number of rules modeled in memory, as shown in Table~\ref{tab:rule counts}. We see that the minimal slice of an intent (the ``Rules in \name" column) is much smaller than that of Libra. Furthermore, this demonstrates the rule duplication issue of Libra as descr... |
2405.21003_TAB_1 | 2405.21003 | https://arxiv.org/abs/2405.21003 | Fidelity, number of rules and coverage for Anchors, GLocalX and CEGA. | The results from comparing the characteristic rules of CEGA to GLocalX and Anchors are summarized in Table~\ref{tab:gloX}. It can be seen that CEGA performs on par with Anchors with respect to fidelity, while producing much fewer unique rules. At the same time, CEGA generally obtains a higher fidelity than GLocalX. |
2405.21003_TAB_3 | 2405.21003 | https://arxiv.org/abs/2405.21003 | The top 11 discriminative rules output by CEGA for the German Credit dataset when using SHAP as the local explainer. | Since CEGA allows for changing the rule format with just a minor modification to the algorithm, as described above, we can easily investigate whether CEGA's improved trade-off between fidelity and complexity (measured by the number of rules) compared to Anchors and GLocalX is due to the use of characteristic rules, or ... |
2405.21003_TAB_4 | 2405.21003 | https://arxiv.org/abs/2405.21003 | Fidelity, number of rules and coverage of characteristic rules output by CEGA using Anchors, SHAP, or LIME | Since CEGA is agnostic to the local explanation technique, it is possible to compare the rules obtained using different local explanation techniques, which allows for evaluating the local explainer on a specific task and selecting the one with the highest fidelity. Accordingly, we compare the fidelity, the number of ru... |
2302.13251_TAB_1 | 2302.13251 | https://arxiv.org/abs/2302.13251 | \textbf{25\% dose level $\rightarrow$ 5\% dose level:} The experimental results on the \texttt{AAPM-A} dataset, \textit{i.e.}, the \texttt{AAPM-16} dataset as the source domain, and the \texttt{AAPM-A} dataset as the target domain. The average value and standard deviation are reported by running each method with fiv... | The quantitative performance of reconstructed results on \texttt{AAPM-A} dataset (as the target domain) can be found in Table \ref{aapm5}. Some observations can be noticed as follows: First, traditional model-based LDCT image reconstruction methods, such as BM3D and ONLM, perform well in terms of PSNR and SSIM, but th... |
2302.13251_TAB_2 | 2302.13251 | https://arxiv.org/abs/2302.13251 | \textbf{25\% dose level $\rightarrow$ 50\% dose level:} The experimental results on the \texttt{AAPM-B} dataset, \textit{i.e.}, the \texttt{AAPM-16} dataset as the source domain, and the \texttt{AAPM-B} dataset as the target domain. The average value and standard deviation are reported by running each method five tim... | The quantitative results of different baseline methods on \texttt{AAPM-B} dataset are reported in Table \ref{aapm50}. As we can see, all methods (except for CCDnet) achieve obvious improvements in terms of PSNR, which reflects the effectiveness of these methods in such milder noise interference. Note that our proposed ... |
2302.13251_TAB_3 | 2302.13251 | https://arxiv.org/abs/2302.13251 | Experimental results on the ISIDCM-20 dataset. ``Mean" denotes the average CT value in the selected ROIs. ``AR" denotes the absolute error between the average CT value of each method and that of referenced NDCT in the selected ROIs. The best and second-best performance in each column is colorized by the \textcolor{red}... | \subsubsection{Performance comparison on clinical dataset} We follow \cite{10081080} to choose the two most representative ROIs for performance comparison on clinical datasets, as shown in Figures \ref{real1} and \ref{real2}. We calculate the average CT Hounsfield units value and corresponding absolute error with refer... |
2302.13251_TAB_4 | 2302.13251 | https://arxiv.org/abs/2302.13251 | Ablation analysis of each component of our proposed method. SDA denotes the sharpness-aware distribution alignment module. BUA denotes the Bayesian uncertainty alignment module. CS and CMC denote the channel-wise statistics and covariance matrix computation in the BUA module, respectively. The best and second-best resu... | The ablation study of our proposed method can be found in Table \ref{ablation study}. As we can see, our proposed Bayesian uncertainty alignment module (Table \hyperlink{ablation study}{V(e)}) takes more improvements in terms of PSNR, GMSD, and DSS, compared with the sharpness-aware distribution alignment module (Table... |
2302.13251_TAB_5 | 2302.13251 | https://arxiv.org/abs/2302.13251 | Ablation study of comparison between CNN-based and transformer-based Bayesian frameworks. For PSNR and SSIM, the higher, the better. For GMSD and DSS, the lower, the better. | \subsubsection{The comparison of different backbone networks} we introduce the \textbf{Hformer}\cite{zhang2023hformer}, a transformer-based LDCT reconstruction model, to play the backbone network of our proposed method. Specifically, we still convert the final layer of the encoder and the decoder of the Hformer to a B... |
2302.13251_TAB_7 | 2302.13251 | https://arxiv.org/abs/2302.13251 | Ablation study of different numbers of Bayesian layers in the encoder and decoder. Note that the structure of the encoder and decoder is symmetric. For PSNR and SSIM, the higher, the better. For GMSD and DSS, the lower, the better. | \subsubsection{The comparison of different Bayesian layers} We present an ablation study in terms of different numbers of Bayesian layers. The results can be found in Table \ref{ablation bayesian layer}. As we can see, the increase in Bayesian layer tasks marginal improvements. However, the computational cost will inc... |
2302.13251_TAB_8 | 2302.13251 | https://arxiv.org/abs/2302.13251 | Subjective Quality Score (Mean ± SD) for Different Methods. The best and second-best performances are bolded and underlined, respectively. | assessment from radiologists’ perspective. The results of the double-blind study can be found in Table \ref{double blind}. As we can see, our proposed method achieves the best or second-best performance among different methods. Note that our proposed method not only has the best overall clinical usability with the bes... |
2303.04501_TAB_1 | 2303.04501 | https://arxiv.org/abs/2303.04501 | \label{t:gaap} Comparison of existing tools used for environmental science {\em (a dash indicates an out-of-scope area)} | We survey existing systems via two strategies: use existing end-to-end platforms that cover as much of the lifecycle we have identified as possible, or pull together a custom system using off-the-shelf components. Our survey is summarized in Table~\ref{t:gaap}. At first glance, a cloud platform like GEE solves the shor... |
2305.06709_TAB_3 | 2305.06709 | https://arxiv.org/abs/2305.06709 | \label{tab:pkg-times} Comparison of different \proglang{Python} packages for Bayesian optimization. The elapsed time per iteration averaged across the ten runs is given for each package. | However, the time \pkg{NUBO} requires to complete one iteration with a maximum of 2.20s for D) is, on average, higher than for the other packages (Table~\ref{tab:pkg-times}). While this might be important for some areas of optimization, it is negligible when optimizing expensive black-box functions, as these functions ... |
2307.10182_TAB_3 | 2307.10182 | https://arxiv.org/abs/2307.10182 | Comparison experiments on different thick-slice simulation methods | The results presented in Table \ref{tab:simulated_TS} offer a detailed comparative analysis of various thick-slice simulation methods applied to two datasets from the 2016 Low Dose CT Grand Challenge, utilizing both the PSNR and the RMSE as the primary performance metrics. We investigated our second hypothesis by train... |
2307.10182_TAB_4 | 2307.10182 | https://arxiv.org/abs/2307.10182 | Multivariate Cox Regression on radiomics features extracted from super-resolved images. | The initial step involved applying ESRGAN models, trained on both naive and our simulation, to these thick-slice images, followed by airway segmentation inference. Subsequently, we filtered out radiomics features with low variance and selected five pertinent features relevant to mortality. The features we analysed incl... |
2307.10182_TAB_5 | 2307.10182 | https://arxiv.org/abs/2307.10182 | The inference time, parameter size and performance of the 3D SR models implemented and refined by our paper. The performance is ranked by the mean value of the PSNR of these models on all testing datasets. | In our research, we conducted comparisons among various models to evaluate their SR effectiveness. The model that emerged as the most effective was the ESRGAN. This highlights the significant role of perceptual loss in SR models. For the benefit of other researchers, we have provided a summary of their efficiencies in ... |
2309.02072_TAB_1 | 2309.02072 | https://arxiv.org/abs/2309.02072 | The estimated parameters of global GARCH(1,1) models trained with different numbers of simulated series. | From Figure \ref{fig:simulation_scaling_garch_10}, it is evident that global training significantly impairs the forecast accuracy of GARCH models, even with just 10 series in the training set. The accuracy of the forecasts diminishes further as the number of training series increases. This is expected because the GARCH... |
2309.02072_TAB_5 | 2309.02072 | https://arxiv.org/abs/2309.02072 | Training configurations for the empirical study. | Similarly to the simulation studies, we trained various global DL models using different numbers of stocks. To investigate how the network architectures respond to the data scaling effect, we repeated the experiments four times for each of RNN, GRU, LSTM, and Transformer. Given the smaller data size compared to the sim... |
2309.10689_TAB_2 | 2309.10689 | https://arxiv.org/abs/2309.10689 | \revadd{We numerically evaluate the effect of reshading in isolation. Our reshading network produces results that are closer to the ground truth than the input.} | Furthermore, we numerically compare our view synthesis results against the other techniques on three synthetic scenes\revdel{(Teapot and Bathroom on top and bottom, respectively)} in Table~\ref{tab:one}. \revadd{To demonstrate that our approach can be used with any pixel relocation method, we show results with both 3D ... |
2309.10689_TAB_3 | 2309.10689 | https://arxiv.org/abs/2309.10689 | We show numerical comparisons against variations of our approach without disparity, frequency encoding, and MLP. \revadd{The results are averaged over the three synthetic scenes}. As shown, all these components are necessary to achieve the best results. | Next, we discuss the effect of several design choices in our approach numerically (Table~\ref{tab:three}) and visually (Figs.~~\ref{fig:ablation_depth},~\ref{fig:ablation_freq_enc},~and~\ref{fig:ablation_mlp}). In Fig.~\ref{fig:ablation_depth}, we demonstrate that without the disparity as the input, our reshading netwo... |
2312.00327_TAB_1 | 2312.00327 | https://arxiv.org/abs/2312.00327 | \centering Determination of order for convergence in Figs. \ref{fig:sphere-time-convergence},\ref{fig:bunny-time-convergence} and \ref{fig:fandisk-time-convergence} | In Figure \ref{fig:space-convergence}, we present an experiment indicating convergence under spatial refinement. While our choice of discretization is standard first-order FEM, we find that the order of convergence of our method under spatial refinement is second-order (see Table \ref{tab:space-convergence}). In Figure... |
2312.04324_TAB_2 | 2312.04324 | https://arxiv.org/abs/2312.04324 | Information about the number of files, the minimum and maximum number of speakers per recording and the number of hours per partition as well as evaluation collar, types of microphone and characteristics of each evaluation dataset. | More information about each dataset can be found in Table~\ref{tab:wideband_datsets_information}. The choice of forgiveness collar for calculating DER corresponds to the least forgiving choice (i.e. collar of 0\,s) except in cases where a challenge or the authors proposed differently. In no case is used any kind of ora... |
2312.04324_TAB_8 | 2312.04324 | https://arxiv.org/abs/2312.04324 | DER (\%) comparison on CH2. For each method (EEND-EDA and DiaPer), selecting the best model on CH1 out of the 5 runs. | & \multirow{2}{*}{Adaptation (multi-speaker SC)} & B & A* & SC2-7 8\,kHz & 75 & Noam & Figures~\ref{fig:telephone_more_speakers}, \ref{fig:confmatrixavg} and Tables~\ref{tab:ch2_per_speaker}, \ref{tab:ch2_der_detail} \\ & & F & B & CH1 & 20 & Adam $10^{-4}$ & Figure~\ref{fig:telephone_more_speakers} and Tabl... |
2312.04324_TAB_9 | 2312.04324 | https://arxiv.org/abs/2312.04324 | Comparison on CH2. For each method, selecting the best model on CH1 out of the 5 runs. DER and its three components and precision and recall for VAD and OSD performance. | & \multirow{2}{*}{Adaptation (multi-speaker SC)} & B & A* & SC2-7 8\,kHz & 75 & Noam & Figures~\ref{fig:telephone_more_speakers}, \ref{fig:confmatrixavg} and Tables~\ref{tab:ch2_per_speaker}, \ref{tab:ch2_der_detail} \\ & & F & B & CH1 & 20 & Adam $10^{-4}$ & Figure~\ref{fig:telephone_more_speakers} and Tabl... |
2312.10819_TAB_1 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Performance metrics for 2020 test set used for assessing accuracy of annual cropland map. We report metrics for other cropland and land cover maps on the same test set for comparison to our map (ordered by decreasing F1 score). Legend: OA = Overall Accuracy, PA = Producer's Accuracy, UA = User's Accuracy, TN = True Neg... | Table \ref{tab:2020-accuracy} and Table \ref{tab:2021-accuracy} report the overall accuracy, precision (user's accuracy), recall (producer's accuracy), and F1 score for the cropland class in the 2020 and 2021 maps, respectively. We computed these metrics using the independent test sets labeled for each year described i... |
2312.10819_TAB_2 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Performance metrics for 2021 test set used for assessing accuracy of annual cropland map. We report metrics for other cropland and land cover maps on the same test set for comparison to our map (ordered by decreasing F1 score). Legend: OA = Overall Accuracy, PA = Producer's Accuracy, UA = User's Accuracy, TN = True Neg... | Table \ref{tab:2020-accuracy} and Table \ref{tab:2021-accuracy} report the overall accuracy, precision (user's accuracy), recall (producer's accuracy), and F1 score for the cropland class in the 2020 and 2021 maps, respectively. We computed these metrics using the independent test sets labeled for each year described i... |
2312.10819_TAB_3 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Change area estimates for Tigray region and each zone within Tigray (units are kha) | Table \ref{tab:change-area} summarizes the area estimates for the four transition classes. We found that cropland gain accounted for $70-160$ kha while cropland loss accounted for $19-79$ kha in Tigray. The stable transition classes accounted for the majority of the area in Tigray: $1,033-1,293$ kha of stable cropland ... |
2312.10819_TAB_5 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Accuracy metrics for four-class cropland change reference sample in Southern zone of Tigray. Overall accuracy is $0.65 \pm 0.04$. Legend: UA = User's Accuracy, PA = Producer's Accuracy. | The change estimates at the zone scale follow a similar trend as the region (Tigray) scale. The majority of the area in each zone was estimated as stable non-crop followed by stable cropland. A relatively small amount of cropland gain and loss was estimated in each zone. However, the standard errors relative to the are... |
2312.10819_TAB_6 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Accuracy metrics for four-class cropland change reference sample in North Western zone of Tigray. Overall accuracy is $0.71 \pm 0.05$. Legend: UA = User's Accuracy, PA = Producer's Accuracy. | The change estimates at the zone scale follow a similar trend as the region (Tigray) scale. The majority of the area in each zone was estimated as stable non-crop followed by stable cropland. A relatively small amount of cropland gain and loss was estimated in each zone. However, the standard errors relative to the are... |
2312.10819_TAB_7 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Accuracy metrics for four-class cropland change reference sample inside 5km buffer of conflict events in Tigray. Overall accuracy is $0.78 \pm 0.05$. Legend: UA = User's Accuracy, PA = Producer's Accuracy. | Table \ref{tab:change-area-buffer} summarizes the change area estimates computed inside and outside of the 5 km conflict buffer. We reported the confidence interval in units of kha and as a percentage of the total area in each boundary to enable comparison between the two boundaries, since the total area outside the bu... |
2312.10819_TAB_8 | 2312.10819 | https://arxiv.org/abs/2312.10819 | Accuracy metrics for four-class cropland change reference sample outside 5km buffer of conflict events in Tigray. Overall accuracy is $0.77 \pm 0.03$. Legend: UA = User's Accuracy, PA = Producer's Accuracy. | Table \ref{tab:change-area-buffer} summarizes the change area estimates computed inside and outside of the 5 km conflict buffer. We reported the confidence interval in units of kha and as a percentage of the total area in each boundary to enable comparison between the two boundaries, since the total area outside the bu... |
2401.01163_TAB_1 | 2401.01163 | https://arxiv.org/abs/2401.01163 | Comparative Properties of Raw and Compressed Videos, Including Metrics such as Bitrate, Frame Rate, and Resolution. | Table \ref{tab1} delineates the attributes of both the compressed, raw videos utilized in our model, with a chosen resolution of 320\texttimes240. This resolution was selected for two pivotal reasons: initially, the resource requirements for training our model with larger frames exceeded our available resources. Subseq... |
2401.05585_TAB_1 | 2401.05585 | https://arxiv.org/abs/2401.05585 | Summary of single-update resilience experiments | \noindent The results are summarized in Table \ref{table:single-update-experiment}. \texttt{UPDATE} is one of the three flight digressions, PROP is goal(tc:TConf) (indicated by \texttt{g}) or \texttt{not(goal(tc:TConf))} (indicated by \texttt{ng}). In both cases, it is required that the single update was applied using ... |
2401.05585_TAB_2 | 2401.05585 | https://arxiv.org/abs/2401.05585 | Summary of simple $n$-resilience experiments | The results are summarized in Table \ref{table:n-resilience-experiment}. Note that \texttt{N} is the number of update applications, and in both PROP cases it is required that \texttt{N} updates were applied (the counter is 0). The remaining parameters and columns are interpreted as in the previous table. |
2402.07688_TAB_2 | 2402.07688 | https://arxiv.org/abs/2402.07688 | Distribution of accuracy among participants. | The \texttt{CyberMetric} dataset covers a diverse range of topics, making it challenging for human experts to recall everything from memory in a closed-book scenario, without prior preparation. The highest score is $71/80\approx 88.75\%$ by an individual holding a PhD and certifications in CISSP, OSCP, and ISO27001LA. ... |
2402.08017_TAB_3 | 2402.08017 | https://arxiv.org/abs/2402.08017 | WER comparison on public wild text benchmarks. \textsc{Lumos} STR obtains the lowest WER with a small size, and the on-device model sacrifices quality only slightly. | It is worth mentioning that placing STR on-device poses significant constraints on the model's architecture, latency, memory, and battery consumption, in addition to the quality challenges for in-the-wild text STR discussed in Section~\ref{sec:intro}. Despite these limitations, our on-device STR model achieves strong ... |
2402.08017_TAB_4 | 2402.08017 | https://arxiv.org/abs/2402.08017 | WER gains from each component | {\bf Ablation study} We now listed the main drivers for WER improvements. We compared with Rosetta, a two-step STR system (faster-rcnn \cite{faster2016} word detector and CNN + CTC recognizer) on the {\em In-house wild text benchmark}. There are three contributors for quality improvements as shown in Table~\ref{tab:att... |
2402.08017_TAB_6 | 2402.08017 | https://arxiv.org/abs/2402.08017 | Model execution metrics. Running the models on hardware accelerator (HA) saved latency by 9X and energy by 3X comparing with running on CPU. | Finally, we show the efficiency of our STR models in Table~\ref{tab:htpcpu} when running on testing devices. The model export steps generated on-device compatible models with the total size around 8Mb. Running the models on hardware accelerator provided huge gain in terms of both latency (9x) and battery usage (3x). |
2402.08017_TAB_7 | 2402.08017 | https://arxiv.org/abs/2402.08017 | Sample prompt to different variants of \textsc{Lumos} for Figure~\ref{fig:prompt_example} | \textbf{Cloud-side:} The cloud side hosts a {\em MM-LLM model}, which takes as input the low-resolution thumbnail, a prompt composed of the recognized texts and their coordinates from STR, and the user query from ASR, and generates the answer response. An illustrative prompt to MM-LLM can be found in Appendix Table \re... |
2403.00376_TAB_1 | 2403.00376 | https://arxiv.org/abs/2403.00376 | Zero-shot classification performance of BLIP-2. | As shown in the Table.~\ref{tab:blip_other}, the performance of BLIP-2 has been improved after applying \eraser, which proves that the objects in the images are discernible, and the misclassifications by the CLIP model are due to decision shortcuts caused by the background. |
2403.00376_TAB_3 | 2403.00376 | https://arxiv.org/abs/2403.00376 | Zero-shot classification performance for different choices of how to construct auxiliary images on real-world scenarios, where ``Average" indicates the averaged performance on all datasets and the \textcolor{darkgreen}{$\blacktriangle$Green} mark represents the improvement relative to Vanilla. | We evaluate the performance of the proposed method on real-world datasets, and the results are presented in Table~\ref{tab:realworld}. ``Patches" denotes the strategy of cropping patches of the test image to serve as auxiliary images, while ``Images" refers to the approach of employing reference images as auxiliary ima... |
2403.00376_TAB_4 | 2403.00376 | https://arxiv.org/abs/2403.00376 | Zero-shot accuracy performance for Vanilla CLIP and InTTA, where ``w/o Anno." indicates the absence of an annotation model like SAM, and the task-irrelevant features are simulated as shown in Fig.~\ref{fig:wosam}. | As shown in Table~\ref{tab:wosam}, even when using randomly perturbed images as spurious features in \eraser, it can still improve (at least without decreasing) CLIP zero-shot classification performance. In other words, our method can still be deployed even without any information that can separate spurious and task-re... |
2403.01964_TAB_7 | 2403.01964 | https://arxiv.org/abs/2403.01964 | Effect of ChatGPT Ban on \texttt{PyPI} Repository Contributors' Output \label{tab:did-repo-lang} | \paragraph{Repository-level Analysis:} For a subset of users working on multiple repositories, we construct a new panel dataset at the user--repository--day level (Section \ref{sec:repo-results} in the appendix).\footnote{Note that the set of public repositories is also restricted to those with at least one user from b... |
2403.01964_TAB_10 | 2403.01964 | https://arxiv.org/abs/2403.01964 | Descriptive Statistics \label{tab:summary-user} | On a daily basis, GitHub actions are relatively rare events at the user level. Hence, we transform each count variable into a binary indicator that equals 1 if one of the actions in a category is recorded for the user on a given day, and 0 otherwise.\footnote{The distribution of day--user-level counts of the main event... |
2403.01964_TAB_11 | 2403.01964 | https://arxiv.org/abs/2403.01964 | Descriptive Statistics \label{tab:summary-user} | On a daily basis, GitHub actions are relatively rare events at the user level. Hence, we transform each count variable into a binary indicator that equals 1 if one of the actions in a category is recorded for the user on a given day, and 0 otherwise.\footnote{The distribution of day--user-level counts of the main event... |
2403.01964_TAB_12 | 2403.01964 | https://arxiv.org/abs/2403.01964 | \label{tab:baseline}The Effect of the ChatGPT Ban on GitHub Output | We then proceed and estimate a difference-in-differences specification that combines the $Pre$ and $Post$ period. The results are presented in Table \ref{tab:baseline}. In the first specification in panel $A$, we specify the pre period as the last Monday and Tuesday (March 27th and 28th 2023) prior to the ban and the p... |
2403.01964_TAB_13 | 2403.01964 | https://arxiv.org/abs/2403.01964 | \label{tab:heterogeneity}Heterogeneity Analysis | Table \ref{tab:heterogeneity} presents the results of a heterogeneity analysis using the sample of users with an organisational affiliation. The results reveal that the adverse effects of the ban on productivity are mainly driven by users who created their profile prior to 2016 and users with 15 followers or less. Whil... |
2403.06512_TAB_1 | 2403.06512 | https://arxiv.org/abs/2403.06512 | Literary Work Applying Threat Modeling to AI Systems | The third and most closely related literature group reports evidence of integrating the AI paradigm within threat modeling. The limited number of publications~\cite{mauri2022modeling} (see \tablename~\ref{tab:rw}) connect potential risks to the elements generated throughout various phases of the life cycle of ML models... |
2403.06512_TAB_3 | 2403.06512 | https://arxiv.org/abs/2403.06512 | Participants Background and Usability Rating | First, the background and expertise of the participants were elicited. As visible from \tablename~\ref{tab:participants}, participants from different backgrounds were selected. None of the participants indicated cybersecurity knowledge, and only two had a Computer Science degree in Data Science. These participants were... |
2403.06512_TAB_5 | 2403.06512 | https://arxiv.org/abs/2403.06512 | Impact Estimates over Ten Years | The impacts were then quantified using the risk simulator in \sol{}. There, the tool gathered the estimated lower and upper bound for each risk, including the probability of occurrence, loss, and confidence level. Two aspects were considered: the applicability of Swiss privacy law with potential fines and the adoption ... |
2403.06512_TAB_6 | 2403.06512 | https://arxiv.org/abs/2403.06512 | Simulated Losses (in CHF) Over Ten Years | After \sol{} performed 100'000 Monte Carlo simulations, the annualized loss expectation was visualized, as summarized in \tablename{}~\ref{tab:exposures}. Based on the range of potential losses, there was high uncertainty regarding the impacts. Furthermore, when interpreting the overall risk exposure (\ie aggregating a... |
2403.09732_TAB_1 | 2403.09732 | https://arxiv.org/abs/2403.09732 | \re{Comparing schema linking (SL) performance (based on table recall metrics)} on Spider dataset between Demo-fixed SL and \texttt{PreSQL}-based SL | Rather than designing strategies such as sorting \cite{c3} or extraction \cite{din-sql} to make LLM output relevant database references, it is better to generate complete \texttt{PreSQL} statements directly. \re{It is also proven in Table \ref{tab:sl_comp}, via such a \texttt{PreSQL} pipeline, our execution accuracy ... |
2403.09732_TAB_2 | 2403.09732 | https://arxiv.org/abs/2403.09732 | The comparison under foundation LLMs on Bird-SQL dataset (9-shot) | Additionally, we have also conducted experiments on the bird develop (bird-dev) dataset \cite{li2024can}, recognized for its heightened complexity. Table \ref{tab:comp_DAIL-SQL_bird} presents the outcomes of SOTA methods with various foundational LLMs. The DAIL-SQL method stands out, exhibiting superior performance acr... |
2403.09732_TAB_3 | 2403.09732 | https://arxiv.org/abs/2403.09732 | The ablation of the $RE_P$ on Bird-dev set (0-shot) | (1) \textbf{\re{Our cell values reference is the most important}}: Discarding it in InternLM and SQLCoder leads to \textbf{-2.4\%} and \textbf{-6.4\%} decreases in EX. This indicates that ensuring the presence of sampled cell values is crucial for LLM generation, and with these references, LLM can better navigate the d... |
2403.09732_TAB_4 | 2403.09732 | https://arxiv.org/abs/2403.09732 | The performance comparison between different prompt styles (0-shot) | We compare our few-shot $RE_p$ with $CR_p$ and $OD_p$, which are recommended by \cite{dail-sql}. It should be noted that the selected question-SQL demonstrations in them are the same. They are evaluated on the Spider development and test datasets separately, with \re{the same selected question-SQL demonstrations in th... |
2404.11343_TAB_2 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Overall model performance (Hit@1) over various datasets. The best performance is denoted in bold. | The results of the recommendation task on four datasets are given in Table \ref{tab: main table}. % In this section, we compare ~\proposed~ to baselines on four datasets. Results are shown in Table \ref{tab: main table}. % , with Movies and TV, and Toys categories experiencing the most pronounced declines. It is worth ... |
2404.11343_TAB_4 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Statistics of the dataset after preprocessing. Avg. Len denotes the average sequence length of users. | \noindent \textbf{Datasets.} For comprehensive evaluations, we used four datasets from Amazon datasets~\cite{mcauley2015image, 10.1145/2872427.2883037}, i.e., Movies and TV, Video Games, Beauty, and Toys, which consist of comprehensive textual information including "title" and "description." Note that we deliberately s... |
2404.11343_TAB_5 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Hyperparameter specifications of ~\proposed | We use the Adam optimizer to train the models in all datasets. For hyperparameters, we tune the model in certain ranges as follows: learning rate $\eta_1, \eta_2$ in $\left\{0.01,0.001, 0.0005, 0.0001\right\}$ for the training stage each, coefficient $\alpha, \beta$ in $\left\{0.1, 0.2, 0.5, 0.75, 1.0\right\}$ for each... |
2404.11343_TAB_6 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Results (Hit@1) on cold/warm item scenario. \proposed~(SBERT) is a variant of \proposed~that uses $\mathbf{q}$ instead of $\mathbf{e}$ for inference. | evaluate cold and warm items in the test set (Table~\ref{tab: cold/warm}). % And measures their performances only regarding the cold/warm items for each scenario. We can conclude the three things via our results in Table \ref{tab: cold/warm}. % Drawing from the results in Table~\ref{tab: cold/warm}, % \textbf{2)} As ... |
2404.11343_TAB_7 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Results (Hit@1) on cold user scenario. | % As shown in Table~\ref{tab: colduser}, we conducted an analysis of model performance, in scenarios where users with a limited interaction history (i.e., cold user) join the platform. Besides evaluations under the cold item scenario, we additionally conduct evaluations under the cold user scenario (Table~\ref{tab: co... |
2404.11343_TAB_8 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Ablation study on Stage-2 of~\proposed~(Hit@1). | Recall that a user representation and item embeddings are injected to the LLM prompt as shown in Figure~\ref{fig: prompt}. In this section, we verify the benefit of injecting them into the prompt (rows (2-4) in Table ~\ref{tab:ablation 2}). We have the following observations: |
2404.11343_TAB_9 | 2404.11343 | https://arxiv.org/abs/2404.11343 | Train/Inference time comparison (Beauty dataset). | As for the training time, we measured the total time spent until the end of training, and as for the inference time, we measured the time spent per mini-batch. Table \ref{tab: model speed} shows that~\proposed~exhibits significantly faster training and inference time compared with TALLRec. Notably, a more substantial i... |
2404.14312_TAB_1 | 2404.14312 | https://arxiv.org/abs/2404.14312 | Overview of the sampling parameters for the training data generation of $T$ samples from the set $B_{M,\tau}^\gamma$ defined in Eq.~\eqref{eq:BMtau}. The same parameters are used for all values of $\gamma$ used in the numerical results section. | We consider test cases in two spatial dimensions so we discard basis elements of $\fm(\fv)$ that correspond to the third spatial dimension. For each closure order $N$, we sample the $B_{M,\tau}^\gamma$, defined in Eq.~\eqref{eq:BMtau}, see Table~\ref{tab_sampling}. $M$ and is dependent on $N$ and chosen such that the ... |
2404.14312_TAB_2 | 2404.14312 | https://arxiv.org/abs/2404.14312 | Architecture and the number of trainable parameters of the neural network models for each closure. A preliminary search is done to determine the a good choice for the network architectures. We report the layer output dimension and the number of layers for each model. ICNNs typically require an order of magnitude fewer ... | We give an overview of the neural network architectures used throughout the numerical results sections in Table~\ref{tab_nn_architecture}. For a given closure, the number of layers and neurons are determined in an architecture search at $\gamma=0$ and then fixed for all regularization levels for comparability, and we ... |
2404.18706_TAB_1 | 2404.18706 | https://arxiv.org/abs/2404.18706 | Results obtained by the image classification model. | The performance results of the model were satisfactory, particularly in terms of minimizing classification ambiguities regarding the \textsc{List} class. The accurate identification of this class is critical, as these pages are subsequently processed by the information extraction model, which requires a high degree of ... |
2405.06270_TAB_3 | 2405.06270 | https://arxiv.org/abs/2405.06270 | Model Performance Metrics for NC Style and NL-Sp Style in \textbf{Zero-shot ICL} setting. Notable performances discussed in the body of the text are highlighted \textbf{in bold}. | \noindent \textbf{Answers to RQ1.} Given the detailed research questions (RQs) and the summary of results from the table provided, an analysis of the performance of LLM-based models in a zero-shot learning setting (RQ1) is structured around two key metrics: accuracy (assessed through F1 score) and risks (assessed throu... |
2405.06270_TAB_4 | 2405.06270 | https://arxiv.org/abs/2405.06270 | Fairness Evaluation for Gender Groups (Top ZeroShot, Bottom Few Shot with 16 examples). Results only focus on the NC style for simplification. | \noindent \textbf{Answer to RQ3.} Table \ref{table:fairness_evaluation} evaluates the fairness of various ML models from a gender perspective, focusing on \RandomForest, \MLPClassifier, \XGBoost, and \LightGBM. Fairness metrics such as recall, precision, and accuracy differences between male (M) and female (F) gender g... |
2405.11431_TAB_5 | 2405.11431 | https://arxiv.org/abs/2405.11431 | Statistical description of the dataset, where the size refers to the number of data points (days). | Due to the incompleteness of data sources, we combine data sources from two websites, including Yahoo Finance\footnote{\url{https://finance.yahoo.com/lookup}} and Kaggle \cite{web:kaggle:cryptoprice} with fundamental details summarised in Table \ref{t411}. The datasets feature the historical price information of the f... |
2405.11431_TAB_7 | 2405.11431 | https://arxiv.org/abs/2405.11431 | Time span of training and test dataset in different experiments for each cryptocurrency. | The data pre-processing in Step 2 features data-scaling to ensure the model stability, which limits the price within the range defined by \textit{min-max} scalar\footnote{\url{https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MinMaxScaler.html}}. Furthermore, we separate the data into 2 sets for c... |
2405.11431_TAB_8 | 2405.11431 | https://arxiv.org/abs/2405.11431 | Hyperparameter selection for the number of hidden units for the selected models using RMSE for the train and test dataset (LSTM, BD-LSTM, ED-LSTM, Conv-LSTM, CNN, and MLP). | We implemented specific experiments to determine the appropriate hyperparameters for each model. Based on related models in the literature \cite{chandra2021evaluation}, we used model architectures: CNN and LSTM variants feature one hidden layer with selected hidden units. We refer to previous research on MLP model for ... |
End of preview. Expand in Data Studio
SciTabData
This dataset contains tables (as .png and .tex files) extracted from scientific articles.
The metadata file contains the caption and referencing text passages for all tables.
To access the raw .png and .tex files, the tar files must be extracted.
Purpose
This dataset is part of the VQA Benchmark developed for my master's thesis. It is designed to evaluate Multimodal LLMs on VQA tasks focused on scientific content.
- Downloads last month
- 534