Distributionally Robust Language Modeling
Oren et al.: https://arxiv.org/abs/1909.02060
#ArtificialIntelligence #DeepLearning #MachineLearning
Oren et al.: https://arxiv.org/abs/1909.02060
#ArtificialIntelligence #DeepLearning #MachineLearning
arXiv.org
Distributionally Robust Language Modeling
Language models are generally trained on data spanning a wide range of topics (e.g., news, reviews, fiction), but they might be applied to an a priori unknown target distribution (e.g., restaurant...
Postdoctoral Position (two years) in Optimization for Statistical Learning
The Department of Mathematics and Mathematical Statistics at Umeå University has an opening for a postdoctoral researcher in mathematical statistics with an emphasis on optimization for statistical learning. The appointment is for two years (subject to satisfactory performance), starting in Fall 2019. The successful candidate is expected to conduct excellent research, actively engage with collaborators, and to participate in the daily activities of the research environment. Last day to apply is September 30, 2019.
Background
The expansion of Artificial Intelligence (AI), in the broad sense, is one of the most exciting developments of the 21st century. This progress opens up many possibilities but also poses grand challenges. The centre Wallenberg AI, Autonomous Systems, and Software Program (WASP) is launching a program to develop the mathematical side of this area. The aim is to strengthen the competence of Sweden as a nation within the area of AI and we are taking part of this program through this specific project. The vision of WASP is excellent research and competence in artificial intelligence, autonomous systems and software for the benefit of Swedish industry. For more information about the research and other activities conducted within WASP, please visit https://wasp-sweden.org/ .
Project description and tasks
Industrial robots, autonomous cars, stocks trading algorithms, and deep network assisted evaluation of medical images all crucially involve real-time, intelligent and automated decision making from complex and heterogeneous data, at ever growing scale and pace. This presents unprecedented theoretical and algorithmic challenges and opportunities in collecting and transforming data into information, predictions and intelligent decisions.
Optimization theory is vital to modern statistical learning and at the forefront of these advances, and the main objective of this postdoctoral position is to develop the next generation of optimization tools to address the above challenges in the context of modern statistical learning, and potentially explore their applications in AI, including medical imaging, automated quality control, and self-driving cars, evaluated on both simulated and real data.
Within this broad framework, the successful candidate is encouraged to develop their own research agenda, in close collaboration with mentors and colleagues. Potential areas of interest include, but not limited to
Training generative adversarial networks,
Nonconvex algorithms for linear inverse problems (such as compressive sensing),
Robust optimization and defense against adversarial examples in deep neural nets,
Role of over-parametrization in training and generalization of deep neural nets,
Global geometry of nonconvex problems,
Efficient and scalable algorithms for constrained nonconvex optimization,
Application of Langevin dynamics and other Monte-Carlo techniques in optimization
Online and storage-optimal algorithms for large scale convex optimization.
For more details and information on how to apply, please visit: https://umu.mynetworkglobal.com/en/what:job/jobID:284855/
The Department of Mathematics and Mathematical Statistics at Umeå University has an opening for a postdoctoral researcher in mathematical statistics with an emphasis on optimization for statistical learning. The appointment is for two years (subject to satisfactory performance), starting in Fall 2019. The successful candidate is expected to conduct excellent research, actively engage with collaborators, and to participate in the daily activities of the research environment. Last day to apply is September 30, 2019.
Background
The expansion of Artificial Intelligence (AI), in the broad sense, is one of the most exciting developments of the 21st century. This progress opens up many possibilities but also poses grand challenges. The centre Wallenberg AI, Autonomous Systems, and Software Program (WASP) is launching a program to develop the mathematical side of this area. The aim is to strengthen the competence of Sweden as a nation within the area of AI and we are taking part of this program through this specific project. The vision of WASP is excellent research and competence in artificial intelligence, autonomous systems and software for the benefit of Swedish industry. For more information about the research and other activities conducted within WASP, please visit https://wasp-sweden.org/ .
Project description and tasks
Industrial robots, autonomous cars, stocks trading algorithms, and deep network assisted evaluation of medical images all crucially involve real-time, intelligent and automated decision making from complex and heterogeneous data, at ever growing scale and pace. This presents unprecedented theoretical and algorithmic challenges and opportunities in collecting and transforming data into information, predictions and intelligent decisions.
Optimization theory is vital to modern statistical learning and at the forefront of these advances, and the main objective of this postdoctoral position is to develop the next generation of optimization tools to address the above challenges in the context of modern statistical learning, and potentially explore their applications in AI, including medical imaging, automated quality control, and self-driving cars, evaluated on both simulated and real data.
Within this broad framework, the successful candidate is encouraged to develop their own research agenda, in close collaboration with mentors and colleagues. Potential areas of interest include, but not limited to
Training generative adversarial networks,
Nonconvex algorithms for linear inverse problems (such as compressive sensing),
Robust optimization and defense against adversarial examples in deep neural nets,
Role of over-parametrization in training and generalization of deep neural nets,
Global geometry of nonconvex problems,
Efficient and scalable algorithms for constrained nonconvex optimization,
Application of Langevin dynamics and other Monte-Carlo techniques in optimization
Online and storage-optimal algorithms for large scale convex optimization.
For more details and information on how to apply, please visit: https://umu.mynetworkglobal.com/en/what:job/jobID:284855/
Varbi
Postdoctoral Position (two years) in Optimization for Statistical Learning
The Department of Mathematics and Mathematical Statistics at Umeå University is opening a postdoctoral position in mathematical statistics within the centre Wallenberg AI, Autonomous Systems, and Software Program (WASP) with an emphasis on optimization for…
Two-year statistics/machine learning postdoctoral position in Sheffield, UK
Dear all,
We are looking to recruit a postdoc for a two-year position in the School of Maths and Statistics and the Department of Computer Science at the University of Sheffield to work on an EPSRC funded project on 'Physically-informed probabilistic modelling of air pollution in Kampala using a low-cost sensor network'.
Closing date: 19 September 2019
Start date: As soon as possible.
https://jobs.shef.ac.uk/sap/bc/webdynpro/sap/hrrcf_a_posting_apply?PARAM=cG9zdF9pbnN0X2d1aWQ9Mzc1MDlCNkFBMDJFMUVEOUIwRTc4MEUxMDcxN0Q5RTkmY2FuZF90eXBlPUVYVA%3d%3d&sap-client=400&sap-language=EN&sap-accessibility=X&sap-ep-themeroot=%2fSAP%2fPUBLIC%2fBC%2fUR%2fuos#
This is a collaborative project between the University of Sheffield and the University of Makerere, and will develop and deploy machine learning methodology to analyse air pollution data from Kampala, in order to determine the source of the pollution and to aid the design of mitigating interventions.
The project team will consist of three Research Associates: two in Kampala and one in Sheffield and will provide methodological support to a larger project funded by a Google Impact Award.
The Sheffield research associate will be supervised by Professor Richard Wilkinson in the School of Mathematics and Statistics and Dr Mauricio Álvarez and Dr Michael Smith in the Department of Computer Science, and will focus on the development of the mathematical tools needed to incorporate physics into machine learning models, and on the development of inferential approaches for these models that are able to deal with large amounts of noisy data. In particular, we aim to develop Gaussian process models that incorporate domain-knowledge about diffusion and advection of air pollution.
We are looking for someone with a PhD in statistics or theoretical/computational physical sciences, ideally with some experience of machine learning.
This is an international multi-party collaboration and will involve travel to Kampala (approximately one trip per year). The project is funded by the EPSRC Global Challenges Research Fund (GCRF), which supports cutting-edge research to address the challenges faced by developing countries.
Best wishes,
Richard and Mauricio
Dear all,
We are looking to recruit a postdoc for a two-year position in the School of Maths and Statistics and the Department of Computer Science at the University of Sheffield to work on an EPSRC funded project on 'Physically-informed probabilistic modelling of air pollution in Kampala using a low-cost sensor network'.
Closing date: 19 September 2019
Start date: As soon as possible.
https://jobs.shef.ac.uk/sap/bc/webdynpro/sap/hrrcf_a_posting_apply?PARAM=cG9zdF9pbnN0X2d1aWQ9Mzc1MDlCNkFBMDJFMUVEOUIwRTc4MEUxMDcxN0Q5RTkmY2FuZF90eXBlPUVYVA%3d%3d&sap-client=400&sap-language=EN&sap-accessibility=X&sap-ep-themeroot=%2fSAP%2fPUBLIC%2fBC%2fUR%2fuos#
This is a collaborative project between the University of Sheffield and the University of Makerere, and will develop and deploy machine learning methodology to analyse air pollution data from Kampala, in order to determine the source of the pollution and to aid the design of mitigating interventions.
The project team will consist of three Research Associates: two in Kampala and one in Sheffield and will provide methodological support to a larger project funded by a Google Impact Award.
The Sheffield research associate will be supervised by Professor Richard Wilkinson in the School of Mathematics and Statistics and Dr Mauricio Álvarez and Dr Michael Smith in the Department of Computer Science, and will focus on the development of the mathematical tools needed to incorporate physics into machine learning models, and on the development of inferential approaches for these models that are able to deal with large amounts of noisy data. In particular, we aim to develop Gaussian process models that incorporate domain-knowledge about diffusion and advection of air pollution.
We are looking for someone with a PhD in statistics or theoretical/computational physical sciences, ideally with some experience of machine learning.
This is an international multi-party collaboration and will involve travel to Kampala (approximately one trip per year). The project is funded by the EPSRC Global Challenges Research Fund (GCRF), which supports cutting-edge research to address the challenges faced by developing countries.
Best wishes,
Richard and Mauricio
Bayesian Inference of Networks Across Multiple Sample Groups and Data Types. https://arxiv.org/abs/1909.02058
arXiv.org
Bayesian Inference of Networks Across Multiple Sample Groups and Data Types
In this paper, we develop a graphical modeling framework for the inference of
networks across multiple sample groups and data types. In medical studies, this
setting arises whenever a set of...
networks across multiple sample groups and data types. In medical studies, this
setting arises whenever a set of...
Minimizing the Societal Cost of Credit Card Fraud with Limited and Imbalanced Data. https://arxiv.org/abs/1909.01486
arXiv.org
Minimizing the Societal Cost of Credit Card Fraud with Limited and...
Machine learning has automated much of financial fraud detection, notifying
firms of, or even blocking, questionable transactions instantly. However, data
imbalance starves traditionally trained...
firms of, or even blocking, questionable transactions instantly. However, data
imbalance starves traditionally trained...
AI Institute "Geometry of Deep Learning" 2019 [Workshop] Day 1 | Session 1
https://www.youtube.com/watch?list=PLD7HFcN7LXRe30qq36It2XCljxc340O_d&v=d9YJJH86ERw&fbclid=IwAR3YzYqnCHznayfiDE1kTH8QFwbljYh7b2v6FAP1_h75rvyEzJSnt_aUdcA&app=desktop
https://www.youtube.com/watch?list=PLD7HFcN7LXRe30qq36It2XCljxc340O_d&v=d9YJJH86ERw&fbclid=IwAR3YzYqnCHznayfiDE1kTH8QFwbljYh7b2v6FAP1_h75rvyEzJSnt_aUdcA&app=desktop
YouTube
AI Institute "Geometry of Deep Learning" 2019 [Workshop] Day 1 | Session 1
Deep learning is transforming the field of artificial intelligence, yet it is lacking solid theoretical underpinnings. This state of affair significantly hin...
rlpyt: A Research Code Base for Deep Reinforcement Learning in PyTorch
Adam Stooke and Pieter Abbeel : https://arxiv.org/abs/1909.01500
#DeepLearning #PyTorch #ReinforcementLearning
Adam Stooke and Pieter Abbeel : https://arxiv.org/abs/1909.01500
#DeepLearning #PyTorch #ReinforcementLearning
arXiv.org
rlpyt: A Research Code Base for Deep Reinforcement Learning in PyTorch
Since the recent advent of deep reinforcement learning for game play and simulated robotic control, a multitude of new algorithms have flourished. Most are model-free algorithms which can be...
From Machine Learning to Machine Reasoning
Bottou et al.: https://arxiv.org/abs/1102.1808
#MachineLearning #MachineReasoning #ArtificialIntelligence
Bottou et al.: https://arxiv.org/abs/1102.1808
#MachineLearning #MachineReasoning #ArtificialIntelligence
arXiv.org
From Machine Learning to Machine Reasoning
A plausible definition of "reasoning" could be "algebraically manipulating previously acquired knowledge in order to answer a new question". This definition covers first-order logical inference or...
[ICCV'19] Code released (https://boqinggong.info/publications.html) for harnessing the potential of simulation for the semantic segmentation of real-world self-driving scenes by using:
Domain generalization: https://arxiv.org/abs/1909.00889,
Domain adaptation: https://arxiv.org/abs/1908.09547,
Thanks to Xiangyu Yue, Yang, and Qing.
Domain generalization: https://arxiv.org/abs/1909.00889,
Domain adaptation: https://arxiv.org/abs/1908.09547,
Thanks to Xiangyu Yue, Yang, and Qing.
boqinggong.info
Publications - Boqing Gong
top paper
Efficient Optimal Planning in non-FIFO Time-Dependent Flow Fields. https://arxiv.org/abs/1909.02198
Efficient Optimal Planning in non-FIFO Time-Dependent Flow Fields. https://arxiv.org/abs/1909.02198
arXiv.org
Efficient Optimal Planning in non-FIFO Time-Dependent Flow Fields
We propose an algorithm for solving the time-dependent shortest path problem
in flow fields where the FIFO (first-in-first-out) assumption is violated. This
problem variant is important for...
in flow fields where the FIFO (first-in-first-out) assumption is violated. This
problem variant is important for...
top paper
Detecting Deep Neural Network Defects with Data Flow Analysis. https://arxiv.org/abs/1909.02190
Detecting Deep Neural Network Defects with Data Flow Analysis. https://arxiv.org/abs/1909.02190
arXiv.org
Detecting Deep Neural Network Defects with Data Flow Analysis
Deep neural networks (DNNs) are shown to be promising solutions in many
challenging artificial intelligence tasks. However, it is very hard to figure
out whether the low precision of a DNN model...
challenging artificial intelligence tasks. However, it is very hard to figure
out whether the low precision of a DNN model...
Poly-GAN: Multi-Conditioned GAN for Fashion Synthesis. https://arxiv.org/abs/1909.02165
Check the final ICCV'19 program here: https://iccv2019.thecvf.com/
It has been a pleasure (and a lot of work) to serve as program chair along with Pr. Svetlana Lazebnik, Pr. Ming-Hsuan Yang and Pr. In So Kweon for the
IEEE/CVF International Conference in #computervision (ICCV'19)
- 4350 full submissions (twice the number of ICCV'17)
- 175 ACs, 1,500 reviewers,
- 13,000 reviews
4 award papers (11 nominations), 200 orals, 850 posters, 25 complaints.
See you in Seoul.
#computervision, #patternrecognition, #artificialintelligence, #machinelearning, #deeplearning
It has been a pleasure (and a lot of work) to serve as program chair along with Pr. Svetlana Lazebnik, Pr. Ming-Hsuan Yang and Pr. In So Kweon for the
IEEE/CVF International Conference in #computervision (ICCV'19)
- 4350 full submissions (twice the number of ICCV'17)
- 175 ACs, 1,500 reviewers,
- 13,000 reviews
4 award papers (11 nominations), 200 orals, 850 posters, 25 complaints.
See you in Seoul.
#computervision, #patternrecognition, #artificialintelligence, #machinelearning, #deeplearning
Here's a list of all the RL papers accepted to NeurIPS 2019!
https://www.endtoend.ai/blog/neurips2019-rl/
https://www.endtoend.ai/blog/neurips2019-rl/
5 Major open problems in NLP
https://deeps.site/blog/2019/09/09/nlp-problems/
Have compiled 5 major problems/opportunities for students, researchers and NLP enthusiasts to work on with open pointers to resources.
https://deeps.site/blog/2019/09/09/nlp-problems/
Have compiled 5 major problems/opportunities for students, researchers and NLP enthusiasts to work on with open pointers to resources.
deeps.site
5 Open problems in NLP
Space to uncover things that tick.
rlpyt: A Research Code Base for Deep Reinforcement Learning in PyTorch
Adam Stooke and Pieter Abbeel : https://arxiv.org/abs/1909.01500
#DeepLearning #PyTorch #ReinforcementLearning
Adam Stooke and Pieter Abbeel : https://arxiv.org/abs/1909.01500
#DeepLearning #PyTorch #ReinforcementLearning
arXiv.org
rlpyt: A Research Code Base for Deep Reinforcement Learning in PyTorch
Since the recent advent of deep reinforcement learning for game play and simulated robotic control, a multitude of new algorithms have flourished. Most are model-free algorithms which can be...
AI-powered banana diseases and pest detection
Paper: https://plantmethods.biomedcentral.com/articles/10.1186/s13007-019-0475-z
These researchers have developed a tool to tackle these silent killers with an app that can scan banana plants for early signs of infection, and alert farmers before it takes hold on their crops.
Paper: https://plantmethods.biomedcentral.com/articles/10.1186/s13007-019-0475-z
These researchers have developed a tool to tackle these silent killers with an app that can scan banana plants for early signs of infection, and alert farmers before it takes hold on their crops.
BioMed Central
AI-powered banana diseases and pest detection - Plant Methods
Background Banana (Musa spp.) is the most popular marketable fruit crop grown all over the world, and a dominant staple food in many developing countries. Worldwide, banana production is affected by numerous diseases and pests. Novel and rapid methods for…
A mathematical model from 103 years ago predicted something that was seen for the first time today: a #black_hole.
#MachineLearning could never do that: it needs observations to model anything. This is a major weak-point of ML. Let's fix it.
A stark contrast between Machine Learning vs other forms of mathematical modeling is that ML models often don't model extreme corner cases very well, because #data in those areas is rare. Gathering data in important areas is as important a skill as building fancy neural networks.
Sadly, too often, using extreme inputs to a model is more useful: e.g. by modeling physics of levers on light objects with short levers, we then built very long levers to lift extremely heavy things. Instead, ML is better suited at modeling everyday phenomena with complex models.
#MachineLearning could never do that: it needs observations to model anything. This is a major weak-point of ML. Let's fix it.
A stark contrast between Machine Learning vs other forms of mathematical modeling is that ML models often don't model extreme corner cases very well, because #data in those areas is rare. Gathering data in important areas is as important a skill as building fancy neural networks.
Sadly, too often, using extreme inputs to a model is more useful: e.g. by modeling physics of levers on light objects with short levers, we then built very long levers to lift extremely heavy things. Instead, ML is better suited at modeling everyday phenomena with complex models.
Blake Richards: Deep Learning with Ensembles of Neocortical Microcircuits
An inspiring talk by Blake Richards at the ICLR2018 emphasizing the interaction between neuroscience and machine learning. This intersection is where great things happen.
https://goo.gl/1YCjrm
https://t.iss.one/ArtificialIntelligenceArticles
An inspiring talk by Blake Richards at the ICLR2018 emphasizing the interaction between neuroscience and machine learning. This intersection is where great things happen.
https://goo.gl/1YCjrm
https://t.iss.one/ArtificialIntelligenceArticles
YouTube
Blake Richards: Deep Learning with Ensembles of Neocortical Microcircuits (ICLR 2018 invited talks)
Abstract: Deep learning in artificial intelligence (AI) has demonstrated that learning hierarchical representations is a good approach for generating useful sensorimotor behaviors. However, the key to effective hierarchical learning is a mechanism for ""credit…