Spontaneous speaking style exhibits notable differences from other speaking styles due to various spontaneous phenomena (e.g., filled pauses, prolongation) and substantial prosody variation (e.g., diverse pitch and duration variation, occasional non-verbal speech like smile), posing challenges to modeling and prediction of spontaneous style. Moreover, the limitation of high-quality spontaneous data constrains spontaneous speech generation for speakers without spontaneous data. To address these problems, we propose SponTTS, a two-stage approach based on bottleneck (BN) features to model and transfer spontaneous style for TTS. In the first stage, we adopt a Conditional Variational Autoencoder (CVAE) to capture spontaneous prosody from a BN feature and involve the spontaneous phenomena by the constraint of spontaneous phenomena embedding prediction loss. Besides, we introduce a flow-based predictor to predict a latent spontaneous style representation from the text, which enriches the prosody and context-specific spontaneous phenomena during inference. In the second stage, we adopt a VITS-like module to transfer the spontaneous style learned in the first stage to target speakers. Experiments demonstrate that SponTTS is effective in modeling spontaneous style and transferring the style to the target speakers, generating spontaneous speech with high naturalness, expressiveness, and speaker similarity. The zero-shot spontaneous style TTS test further verifies the generalization and robustness of SponTTS in generating spontaneous speech for unseen speakers.
{"title":"SponTTS: modeling and transferring spontaneous style for TTS","authors":"Li, Hanzhao, Zhu, Xinfa, Xue, Liumeng, Song, Yang, Chen, Yunlin, Xie, Lei","doi":"10.48550/arxiv.2311.07179","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07179","url":null,"abstract":"Spontaneous speaking style exhibits notable differences from other speaking styles due to various spontaneous phenomena (e.g., filled pauses, prolongation) and substantial prosody variation (e.g., diverse pitch and duration variation, occasional non-verbal speech like smile), posing challenges to modeling and prediction of spontaneous style. Moreover, the limitation of high-quality spontaneous data constrains spontaneous speech generation for speakers without spontaneous data. To address these problems, we propose SponTTS, a two-stage approach based on bottleneck (BN) features to model and transfer spontaneous style for TTS. In the first stage, we adopt a Conditional Variational Autoencoder (CVAE) to capture spontaneous prosody from a BN feature and involve the spontaneous phenomena by the constraint of spontaneous phenomena embedding prediction loss. Besides, we introduce a flow-based predictor to predict a latent spontaneous style representation from the text, which enriches the prosody and context-specific spontaneous phenomena during inference. In the second stage, we adopt a VITS-like module to transfer the spontaneous style learned in the first stage to target speakers. Experiments demonstrate that SponTTS is effective in modeling spontaneous style and transferring the style to the target speakers, generating spontaneous speech with high naturalness, expressiveness, and speaker similarity. The zero-shot spontaneous style TTS test further verifies the generalization and robustness of SponTTS in generating spontaneous speech for unseen speakers.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"118 15","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353280","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Developing novel high-temperature van der Waals ferromagnetic semiconductor materials and investigating their interface coupling effects with two-dimensional topological semimetals are pivotal for advancing next-generation spintronic and quantum devices. However, most van der Waals ferromagnetic semiconductors exhibit ferromagnetism only at low temperatures, limiting the proximity research on their interfaces with topological semimetals. Here, we report an intrinsic, van der Waals layered room-temperature ferromagnetic semiconductor crystal, FeCr0.5Ga1.5Se4 (FCGS), with a Curie temperature as high as 370 K, setting a new record for van der Waals ferromagnetic semiconductors. The saturation magnetization at low temperature (2 K) and room temperature (300 K) reaches 8.2 emu/g and 2.7 emu/g, respectively. Furthermore, FCGS possesses a bandgap of approximately 1.2 eV, which is comparable to the widely used commercial silicon. The FCGS/graphene heterostructure exhibits an impeccably smooth and gapless interface, thereby inducing a robust magnetic proximity coupling effect between FCGS and graphene. After the proximity coupling, graphene undergoes a charge carrier transition from electrons to holes, accompanied by a transition from non-magnetic to ferromagnetic transport behavior with robust anomalous Hall effect. Notably, the anomalous Hall effect remains robust even temperatures up to 400 K.
{"title":"Robust magnetic proximity induced anomalous Hall effect in a room\u0000 temperature van der Waals ferromagnetic semiconductor based 2D\u0000 heterostructure","authors":"Wu, Hao, Yang, Li, Zhang, Gaojie, Jin, Wen, Xiao, Bichen, Zhang, Wenfeng, Chang, Haixin","doi":"10.48550/arxiv.2311.07183","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07183","url":null,"abstract":"Developing novel high-temperature van der Waals ferromagnetic semiconductor materials and investigating their interface coupling effects with two-dimensional topological semimetals are pivotal for advancing next-generation spintronic and quantum devices. However, most van der Waals ferromagnetic semiconductors exhibit ferromagnetism only at low temperatures, limiting the proximity research on their interfaces with topological semimetals. Here, we report an intrinsic, van der Waals layered room-temperature ferromagnetic semiconductor crystal, FeCr0.5Ga1.5Se4 (FCGS), with a Curie temperature as high as 370 K, setting a new record for van der Waals ferromagnetic semiconductors. The saturation magnetization at low temperature (2 K) and room temperature (300 K) reaches 8.2 emu/g and 2.7 emu/g, respectively. Furthermore, FCGS possesses a bandgap of approximately 1.2 eV, which is comparable to the widely used commercial silicon. The FCGS/graphene heterostructure exhibits an impeccably smooth and gapless interface, thereby inducing a robust magnetic proximity coupling effect between FCGS and graphene. After the proximity coupling, graphene undergoes a charge carrier transition from electrons to holes, accompanied by a transition from non-magnetic to ferromagnetic transport behavior with robust anomalous Hall effect. Notably, the anomalous Hall effect remains robust even temperatures up to 400 K.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"118 6","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353289","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07342
Bernardi, Olga, Florio, Anna, Leguil, Martin
We study the dynamics of dissipative billiard maps within planar convex domains. Such maps have a global attractor. We are interested in the topological and dynamical complexity of the attractor, in terms both of the geometry of the billiard table and of the strength of the dissipation. We focus on the study of an invariant subset of the attractor, the so-called Birkhoff attractor. On the one hand, we show that for a generic convex table with "pinched" curvature, the Birkhoff attractor is a normally contracted manifold when the dissipation is strong. On the other hand, for a mild dissipation, we prove that generically the Birkhoff attractor is complicated, both from the topological and the dynamical point of view.
{"title":"Birkhoff attractors of dissipative billiards","authors":"Bernardi, Olga, Florio, Anna, Leguil, Martin","doi":"10.48550/arxiv.2311.07342","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07342","url":null,"abstract":"We study the dynamics of dissipative billiard maps within planar convex domains. Such maps have a global attractor. We are interested in the topological and dynamical complexity of the attractor, in terms both of the geometry of the billiard table and of the strength of the dissipation. We focus on the study of an invariant subset of the attractor, the so-called Birkhoff attractor. On the one hand, we show that for a generic convex table with \"pinched\" curvature, the Birkhoff attractor is a normally contracted manifold when the dissipation is strong. On the other hand, for a mild dissipation, we prove that generically the Birkhoff attractor is complicated, both from the topological and the dynamical point of view.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"114 3","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353292","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07499
Zhang, Xiang, Tomizuka, Masayoshi, Li, Hui
Contact-rich manipulation tasks often exhibit a large sim-to-real gap. For instance, industrial assembly tasks frequently involve tight insertions where the clearance is less than (0.1) mm and can even be negative when dealing with a deformable receptacle. This narrow clearance leads to complex contact dynamics that are difficult to model accurately in simulation, making it challenging to transfer simulation-learned policies to real-world robots. In this paper, we propose a novel framework for robustly learning manipulation skills for real-world tasks using only the simulated data. Our framework consists of two main components: the ``Force Planner'' and the ``Gain Tuner''. The Force Planner is responsible for planning both the robot motion and desired contact forces, while the Gain Tuner dynamically adjusts the compliance control gains to accurately track the desired contact forces during task execution. The key insight of this work is that by adaptively adjusting the robot's compliance control gains during task execution, we can modulate contact forces in the new environment, thereby generating trajectories similar to those trained in simulation and narrows the sim-to-real gap. Experimental results show that our method, trained in simulation on a generic square peg-and-hole task, can generalize to a variety of real-world insertion tasks involving narrow or even negative clearances, all without requiring any fine-tuning.
{"title":"Bridging the Sim-to-Real Gap with Dynamic Compliance Tuning for\u0000 Industrial Insertion","authors":"Zhang, Xiang, Tomizuka, Masayoshi, Li, Hui","doi":"10.48550/arxiv.2311.07499","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07499","url":null,"abstract":"Contact-rich manipulation tasks often exhibit a large sim-to-real gap. For instance, industrial assembly tasks frequently involve tight insertions where the clearance is less than (0.1) mm and can even be negative when dealing with a deformable receptacle. This narrow clearance leads to complex contact dynamics that are difficult to model accurately in simulation, making it challenging to transfer simulation-learned policies to real-world robots. In this paper, we propose a novel framework for robustly learning manipulation skills for real-world tasks using only the simulated data. Our framework consists of two main components: the ``Force Planner'' and the ``Gain Tuner''. The Force Planner is responsible for planning both the robot motion and desired contact forces, while the Gain Tuner dynamically adjusts the compliance control gains to accurately track the desired contact forces during task execution. The key insight of this work is that by adaptively adjusting the robot's compliance control gains during task execution, we can modulate contact forces in the new environment, thereby generating trajectories similar to those trained in simulation and narrows the sim-to-real gap. Experimental results show that our method, trained in simulation on a generic square peg-and-hole task, can generalize to a variety of real-world insertion tasks involving narrow or even negative clearances, all without requiring any fine-tuning.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"108 4","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353313","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We introduce SPUD (Semantically Perturbed Universal Dependencies), a framework for creating nonce treebanks for the multilingual Universal Dependencies (UD) corpora. SPUD data satisfies syntactic argument structure, provides syntactic annotations, and ensures grammaticality via language-specific rules. We create nonce data in Arabic, English, French, German, and Russian, and demonstrate two use cases of SPUD treebanks. First, we investigate the effect of nonce data on word co-occurrence statistics, as measured by perplexity scores of autoregressive (ALM) and masked language models (MLM). We find that ALM scores are significantly more affected by nonce data than MLM scores. Second, we show how nonce data affects the performance of syntactic dependency probes. We replicate the findings of M"uller-Eberstein et al. (2022) on nonce test data and show that the performance declines on both MLMs and ALMs wrt. original test data. However, a majority of the performance is kept, suggesting that the probe indeed learns syntax independently from semantics.
{"title":"Multilingual Nonce Dependency Treebanks: Understanding how LLMs\u0000 represent and process syntactic structure","authors":"Arps, David, Kallmeyer, Laura, Samih, Younes, Sajjad, Hassan","doi":"10.48550/arxiv.2311.07497","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07497","url":null,"abstract":"We introduce SPUD (Semantically Perturbed Universal Dependencies), a framework for creating nonce treebanks for the multilingual Universal Dependencies (UD) corpora. SPUD data satisfies syntactic argument structure, provides syntactic annotations, and ensures grammaticality via language-specific rules. We create nonce data in Arabic, English, French, German, and Russian, and demonstrate two use cases of SPUD treebanks. First, we investigate the effect of nonce data on word co-occurrence statistics, as measured by perplexity scores of autoregressive (ALM) and masked language models (MLM). We find that ALM scores are significantly more affected by nonce data than MLM scores. Second, we show how nonce data affects the performance of syntactic dependency probes. We replicate the findings of M\"uller-Eberstein et al. (2022) on nonce test data and show that the performance declines on both MLMs and ALMs wrt. original test data. However, a majority of the performance is kept, suggesting that the probe indeed learns syntax independently from semantics.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"107 23","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353320","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07143
Nguyen, Tien Dat, Kim, Jinwoo, Yang, Hongseok, Hong, Seunghoon
We present a general framework for symmetrizing an arbitrary neural-network architecture and making it equivariant with respect to a given group. We build upon the proposals of Kim et al. (2023); Kaba et al. (2023) for symmetrization, and improve them by replacing their conversion of neural features into group representations, with an optimization whose loss intuitively measures the distance between group orbits. This change makes our approach applicable to a broader range of matrix groups, such as the Lorentz group O(1, 3), than these two proposals. We experimentally show our method's competitiveness on the SO(2) image classification task, and also its increased generality on the task with O(1, 3). Our implementation will be made accessible at https://github.com/tiendatnguyen-vision/Orbit-symmetrize.
{"title":"Learning Symmetrization for Equivariance with Orbit Distance\u0000 Minimization","authors":"Nguyen, Tien Dat, Kim, Jinwoo, Yang, Hongseok, Hong, Seunghoon","doi":"10.48550/arxiv.2311.07143","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07143","url":null,"abstract":"We present a general framework for symmetrizing an arbitrary neural-network architecture and making it equivariant with respect to a given group. We build upon the proposals of Kim et al. (2023); Kaba et al. (2023) for symmetrization, and improve them by replacing their conversion of neural features into group representations, with an optimization whose loss intuitively measures the distance between group orbits. This change makes our approach applicable to a broader range of matrix groups, such as the Lorentz group O(1, 3), than these two proposals. We experimentally show our method's competitiveness on the SO(2) image classification task, and also its increased generality on the task with O(1, 3). Our implementation will be made accessible at https://github.com/tiendatnguyen-vision/Orbit-symmetrize.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"118 43","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353428","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07193
Huey, Greg
Presented are toy-models for sub-luminal and super-luminal warp-drives in 3+1 dimensions. The models are constructed in a chimeric manner - as different bulk space-times separated by thin membranes. The membranes contain perfect-fluid-like stress-energy tensors. The Israel junction conditions relate this stress-energy to a jump in extrinsic curvature across the brane, which in turn manifests as apparent acceleration in the bulk space-times. The acceleration on either side of the brane may be set individually by choice of model parameters. The Weak Energy Condition (WEC) is shown to be satisfied everywhere in both models. Although the branes in these toy models are not compact, it is demonstrated that super-luminal warp-drive is possible that satisfies the WEC. Additionally, the nature of these models provides framework for speculation on a mechanism for transition from sub-luminal to super-luminal warp. Neither quantum effects nor stability of the models is considered.
{"title":"Membrane Models as a Means of Propulsion in General Relativity:\u0000 Super-Luminal Warp-Drive that Satisfies the Weak Energy Condition","authors":"Huey, Greg","doi":"10.48550/arxiv.2311.07193","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07193","url":null,"abstract":"Presented are toy-models for sub-luminal and super-luminal warp-drives in 3+1 dimensions. The models are constructed in a chimeric manner - as different bulk space-times separated by thin membranes. The membranes contain perfect-fluid-like stress-energy tensors. The Israel junction conditions relate this stress-energy to a jump in extrinsic curvature across the brane, which in turn manifests as apparent acceleration in the bulk space-times. The acceleration on either side of the brane may be set individually by choice of model parameters. The Weak Energy Condition (WEC) is shown to be satisfied everywhere in both models. Although the branes in these toy models are not compact, it is demonstrated that super-luminal warp-drive is possible that satisfies the WEC. Additionally, the nature of these models provides framework for speculation on a mechanism for transition from sub-luminal to super-luminal warp. Neither quantum effects nor stability of the models is considered.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"118 33","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353438","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07160
Soto-Sanfiel, María T., Chong, Chin-Wen, Latorre, José I.
An interpretive phenomenological approach is adopted to investigate scientists' attitudes and practices related to hype in science communication. Twenty-four active quantum physicists participated in 5 focus groups. Through a semi-structured questionnaire, their use of hype, attitudes, behaviours, and perspectives on hype in science communication were observed. The main results show that scientists primarily attribute hype generation to themselves, major corporations, and marketing departments. They see hype as crucial for research funding and use it strategically, despite concerns. Scientists view hype as coercive, compromising their work's integrity, leading to mostly negative feelings about it, except for collaborator-generated hype. A dissonance exists between scientists' involvement in hype, their opinions, and the negative emotions it triggers. They manage this by attributing responsibility to the academic system, downplaying their practices. This reveals hype in science communication as a calculated, persuasive tactic by academic stakeholders, aligning with a neoliberal view of science. Implications extend to science communication, media studies, regulation, and academia.
{"title":"Hype in Science Communication: Exploring Scientists' Attitudes and\u0000 Practices in Quantum Physics","authors":"Soto-Sanfiel, María T., Chong, Chin-Wen, Latorre, José I.","doi":"10.48550/arxiv.2311.07160","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07160","url":null,"abstract":"An interpretive phenomenological approach is adopted to investigate scientists' attitudes and practices related to hype in science communication. Twenty-four active quantum physicists participated in 5 focus groups. Through a semi-structured questionnaire, their use of hype, attitudes, behaviours, and perspectives on hype in science communication were observed. The main results show that scientists primarily attribute hype generation to themselves, major corporations, and marketing departments. They see hype as crucial for research funding and use it strategically, despite concerns. Scientists view hype as coercive, compromising their work's integrity, leading to mostly negative feelings about it, except for collaborator-generated hype. A dissonance exists between scientists' involvement in hype, their opinions, and the negative emotions it triggers. They manage this by attributing responsibility to the academic system, downplaying their practices. This reveals hype in science communication as a calculated, persuasive tactic by academic stakeholders, aligning with a neoliberal view of science. Implications extend to science communication, media studies, regulation, and academia.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"114 20","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353449","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07162
An, Taegun, Joo, Changhee
We develop a Neural Architecture Search (NAS) framework for CycleGAN that carries out unpaired image-to-image translation task. Extending previous NAS techniques for Generative Adversarial Networks (GANs) to CycleGAN is not straightforward due to the task difference and greater search space. We design architectures that consist of a stack of simple ResNet-based cells and develop a search method that effectively explore the large search space. We show that our framework, called CycleGANAS, not only effectively discovers high-performance architectures that either match or surpass the performance of the original CycleGAN, but also successfully address the data imbalance by individual architecture search for each translation direction. To our best knowledge, it is the first NAS result for CycleGAN and shed light on NAS for more complex structures.
{"title":"CycleGANAS: Differentiable Neural Architecture Search for CycleGAN","authors":"An, Taegun, Joo, Changhee","doi":"10.48550/arxiv.2311.07162","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07162","url":null,"abstract":"We develop a Neural Architecture Search (NAS) framework for CycleGAN that carries out unpaired image-to-image translation task. Extending previous NAS techniques for Generative Adversarial Networks (GANs) to CycleGAN is not straightforward due to the task difference and greater search space. We design architectures that consist of a stack of simple ResNet-based cells and develop a search method that effectively explore the large search space. We show that our framework, called CycleGANAS, not only effectively discovers high-performance architectures that either match or surpass the performance of the original CycleGAN, but also successfully address the data imbalance by individual architecture search for each translation direction. To our best knowledge, it is the first NAS result for CycleGAN and shed light on NAS for more complex structures.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"114 15","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353454","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2023-11-13DOI: 10.48550/arxiv.2311.07161
Miranda, Lester James V.
We present the development of a Named Entity Recognition (NER) dataset for Tagalog. This corpus helps fill the resource gap present in Philippine languages today, where NER resources are scarce. The texts were obtained from a pretraining corpora containing news reports, and were labeled by native speakers in an iterative fashion. The resulting dataset contains ~7.8k documents across three entity types: Person, Organization, and Location. The inter-annotator agreement, as measured by Cohen's $kappa$, is 0.81. We also conducted extensive empirical evaluation of state-of-the-art methods across supervised and transfer learning settings. Finally, we released the data and processing code publicly to inspire future work on Tagalog NLP.
{"title":"Developing a Named Entity Recognition Dataset for Tagalog","authors":"Miranda, Lester James V.","doi":"10.48550/arxiv.2311.07161","DOIUrl":"https://doi.org/10.48550/arxiv.2311.07161","url":null,"abstract":"We present the development of a Named Entity Recognition (NER) dataset for Tagalog. This corpus helps fill the resource gap present in Philippine languages today, where NER resources are scarce. The texts were obtained from a pretraining corpora containing news reports, and were labeled by native speakers in an iterative fashion. The resulting dataset contains ~7.8k documents across three entity types: Person, Organization, and Location. The inter-annotator agreement, as measured by Cohen's $kappa$, is 0.81. We also conducted extensive empirical evaluation of state-of-the-art methods across supervised and transfer learning settings. Finally, we released the data and processing code publicly to inspire future work on Tagalog NLP.","PeriodicalId":496270,"journal":{"name":"arXiv (Cornell University)","volume":"114 12","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"136353457","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：