Filters








48 Hits in 2.7 sec

TAPAS: Train-less Accuracy Predictor for Architecture Search [article]

R. Istrate, F. Scheidegger, G. Mariani, D. Nikolopoulos, C. Bekas, A. C. I. Malossi
2018 arXiv   pre-print
We present results of two searches performed in 400 seconds on a single GPU. Our best discovered networks reach 93.67% accuracy for CIFAR-10 and 81.01% for CIFAR-100, verified by training.  ...  We propose a new deep neural network accuracy predictor, that estimates in fractions of a second classification performance for unseen input datasets, without training.  ...  In this work, we introduce a train-less accuracy predictor for architecture search (TAPAS), that provides reliable architecture peak accuracy predictions when used with unseen (i.e., not previously seen  ... 
arXiv:1806.00250v1 fatcat:ki5fgkaakjbgvbwxjifalp337m

TAPAS: Train-Less Accuracy Predictor for Architecture Search

R. Istrate, F. Scheidegger, G. Mariani, D. Nikolopoulos, C. Bekas, A. C. I. Malossi
2019 PROCEEDINGS OF THE THIRTIETH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE AND THE TWENTY-EIGHTH INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE CONFERENCE  
We present results of two searches performed in 400 seconds on a single GPU. Our best discovered networks reach 93.67% accuracy for CIFAR-10 and 81.01% for CIFAR-100, verified by training.  ...  We propose a new deep neural network accuracy predictor, that estimates in fractions of a second classification performance for unseen input datasets, without training.  ...  In this work, we introduce a train-less accuracy predictor for architecture search (TAPAS), that provides reliable architecture peak accuracy predictions when used with unseen (i.e., not previously seen  ... 
doi:10.1609/aaai.v33i01.33013927 fatcat:j74brxqefjdkvfybk5ij3jvrxa

Efficient image dataset classification difficulty estimation for predicting deep-learning accuracy

Florian Scheidegger, Roxana Istrate, Giovanni Mariani, Luca Benini, Costas Bekas, Cristiano Malossi
2020 The Visual Computer  
Dario Garcia Gasulla from the Barcelona Supercomputing Center for discussion and advise.  ...  We recently proposed a method called train-less accuracy predictor for architecture search (TAPAS) that demonstrates how the probe nets contribute to bias and perform an architecture search.  ...  Figure 11 shows the high level system that is invoked for the train-less architecture search of TAPAS.  ... 
doi:10.1007/s00371-020-01922-5 fatcat:xii7x6elpbf7jg3fvxxitr4hpm

NeuNetS: An Automated Synthesis Engine for Neural Network Design [article]

Atin Sood, Benjamin Elder, Benjamin Herta, Chao Xue, Costas Bekas, A. Cristiano I. Malossi, Debashish Saha, Florian Scheidegger, Ganesh Venkataraman, Gegi Thomas, Giovanni Mariani, Hendrik Strobelt (+8 others)
2019 arXiv   pre-print
Pre-trained neural network models available through APIs or capability to custom train pre-built neural network architectures with customer data has made the consumption of AI by developers much simpler  ...  NeuNetS is available for both Text and Image domains and can build neural networks for specific tasks in a fraction of the time it takes today with human effort, and with accuracy similar to that of human-designed  ...  Train-less Accuracy Predictor (TAP): Given an NN architecture and a DCN, it predicts the potentially reachable peak accuracy without training the network.  ... 
arXiv:1901.06261v1 fatcat:w4e4celkwfawpdu7vlwgcnzrqe

BANANAS: Bayesian Optimization with Neural Architectures for Neural Architecture Search [article]

Colin White, Willie Neiswanger, Yash Savani
2020 arXiv   pre-print
Over the past half-decade, many methods have been considered for neural architecture search (NAS).  ...  In this work, we give a thorough analysis of the "BO + neural predictor" framework by identifying five main components: the architecture encoding, neural predictor, uncertainty calibration method, acquisition  ...  Acknowledgments We thank Jeff Schneider, Naveen Sundar Govindarajulu, and Liam Li for their help with this project.  ... 
arXiv:1910.11858v3 fatcat:pgrwhrstw5fjtoce2xayohydvq

Constrained deep neural network architecture search for IoT devices accounting hardware calibration [article]

Florian Scheidegger, Luca Benini, Costas Bekas, Cristiano Malossi
2019 arXiv   pre-print
searches.  ...  However, the design of network topologies is a complex task and the research community makes a constant effort in discovering top-accuracy topologies, either manually or employing expensive architecture  ...  A method called train Train-less Accuracy Predictor for Architecture Search (TAPAS) demonstrates how to generalize architecture search results to new data without the need of training during the search  ... 
arXiv:1909.10818v1 fatcat:zoauans5tja2zbh3x3ppwlzd7i

"You might also like this model": Data Driven Approach for Recommending Deep Learning Models for Unknown Image Datasets [article]

Ameya Prabhu, Riddhiman Dasgupta, Anush Sankaran, Srikanth Tamilselvam, Senthil Mani
2020 arXiv   pre-print
Further, we predict the performance accuracy of the recommended architecture on the given unknown dataset, without the need for training the model.  ...  For an unknown (new) classification dataset, choosing an appropriate deep learning architecture is often a recursive, time-taking, and laborious process.  ...  TAPAS [12] is another novel deep neural network accuracy predictor, parameterized on network topology as well as a measure of dataset difficulty. Scheidegger et al.  ... 
arXiv:1911.11433v2 fatcat:w4ki22d2hzhhvf5fy3e4wsqiye

NAS-Bench-301 and the Case for Surrogate Benchmarks for Neural Architecture Search [article]

Julien Siems, Lucas Zimmer, Arber Zela, Jovita Lukasik, Margret Keuper, Frank Hutter
2020 arXiv   pre-print
The most significant barrier to the advancement of Neural Architecture Search (NAS) is its demand for large computational resources, which hinders scientifically sound empirical evaluations.  ...  To overcome this fundamental limitation, we propose NAS-Bench-301, the first surrogate NAS benchmark, using a search space containing 10^18 architectures, many orders of magnitude larger than any previous  ...  Tapas: Train-less accuracy predictor for architecture search. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 33, pp. 3927-3934, 2019. G. Ke, Q. Meng, T. Finley, T. Wang, W.  ... 
arXiv:2008.09777v3 fatcat:auwetqgtdva5li5l2x5skuclka

Understanding tables with intermediate pre-training [article]

Julian Martin Eisenschlos, Syrine Krichene, Thomas Müller
2020 arXiv   pre-print
accuracy.  ...  While there is extensive work on textual entailment, table entailment is less well studied. We adapt TAPAS (Herzig et al., 2020), a table-based BERT model, to recognize entailment.  ...  Acknowledgments We would like to thank Jordan Boyd-Graber, Yasemin Altun, Emily Pitler, Benjamin Boerschinger, William Cohen, Jonathan Herzig, Slav Petrov, and the anonymous reviewers for their time, constructive  ... 
arXiv:2010.00571v2 fatcat:wnjbhlhd4bgobdz3vwpjlbob5q

SpreadsheetCoder: Formula Prediction from Semi-structured Context [article]

Xinyun Chen, Petros Maniatis, Rishabh Singh, Charles Sutton, Hanjun Dai, Max Lin, Denny Zhou
2021 arXiv   pre-print
We train our model on a large dataset of spreadsheets, and demonstrate that SpreadsheetCoder achieves top-1 prediction accuracy of 42.51%, which is a considerable improvement over baselines that do not  ...  In particular, we propose SpreadsheetCoder, a BERT-based model architecture to represent the tabular context in both row-based and column-based formats.  ...  Second, both TAPAS and Table-BERT require that the maximum table size is 512 tokens, which is not enough for our problem.  ... 
arXiv:2106.15339v1 fatcat:nomgidezvrcvbo2b75tjm7ypsq

A Constructive Prediction of the Generalization Error Across Scales [article]

Jonathan S. Rosenfeld, Amir Rosenfeld, Yonatan Belinkov, Nir Shavit
2019 arXiv   pre-print
The dependency of the generalization error of neural networks on model and dataset size is of critical importance both in practice and for understanding the theory of neural networks.  ...  Tapas: Train-less accuracy predictor for architecture search. In Proceed- ings of the AAAI Conference on Artificial Intelligence, volume 33, pp. 3927-3934, 2019.  ...  As figure 3 shows, estimated test accuracy is highly correlated with actual test accuracy for various datasets, with worst-case values µ < 1% and σ < 5% .  ... 
arXiv:1909.12673v2 fatcat:kcrsitz5erg7vaibezdgbdwq2m

Multi-Instance Training for Question Answering Across Table and Linked Text [article]

Vishwajeet Kumar, Saneem Chemmengath, Yash Gupta, Jaydeep Sen, Samarth Bharadwaj, Soumen Chakrabarti
2021 arXiv   pre-print
Training such systems is further challenged by the need for distant supervision.  ...  Existing adaptations of table representation to transformer-based reading comprehension (RC) architectures fail to tackle the diverse modalities of the two representations through a single system.  ...  These weights can either be optimized for using grid search or even trained for using model outputs as features and evaluation scores as labels.  ... 
arXiv:2112.07337v1 fatcat:nqbhzovaqvg6pbvlvkwp3tmbti

Making Table Understanding Work in Practice [article]

Madelon Hulsebos and Sneha Gathani and James Gale and Isil Dillig and Paul Groth and Çağatay Demiralp
2021 arXiv   pre-print
Understanding the semantics of tables at scale is crucial for tasks like data integration, preparation, and search.  ...  With the rise of deep learning, powerful models have been developed for these tasks with excellent accuracy on benchmarks.  ...  as weak predictors in the local model.  ... 
arXiv:2109.05173v1 fatcat:huoeikenzbesbco3ymnd5pxcry

Evaluation of Load Prediction Techniques for Distributed Stream Processing [article]

Kordian Gontarska, Morgan Geldenhuys, Dominik Scheinert, Philipp Wiesner, Andreas Polze, Lauritz Thamsen
2021 arXiv   pre-print
We compare model performance with respect to overall accuracy and training duration.  ...  We identify three use-cases and formulate requirements for making load predictions specific to DSP jobs.  ...  ACKNOWLEDGMENTS This work has been supported through grants by the German Federal Ministry for Economic Affairs and Energy (BMWi) as Telemed5000 (funding mark 01MD19014C), and by the German Federal Ministry  ... 
arXiv:2108.04749v1 fatcat:tl5crw3oezgl5f2wcmdsvpmwqq

Estimating T_ eff, radius and luminosity of M-dwarfs using high resolution optical and NIR spectral features [article]

Dhrimadri Khata, Soumen Mondal, Ramkrishna Das, Tapas Baug
2021 pre-print
We estimate effective temperature (T_ eff), stellar radius, and luminosity for a sample of 271 M-dwarf stars (M0V-M7V) observed as a part of CARMENES (Calar Alto high-Resolution search for M dwarfs with  ...  We also explore and compare our results with literature values obtained using other different methods for the same sample of M dwarfs.  ...  More recently machine learning techniques have been applied widely, where different neural network architectures are built and trained to determine stellar parameters of M dwarfs (Sarro et al. 2018 ,  ... 
doi:10.1093/mnras/stab2211 arXiv:2107.14023v1 fatcat:ik7kcyrdzjfcxkogxjosvh4zs4
« Previous Showing results 1 — 15 out of 48 results