Announcement

Date and time: Dec 09, 2016, 11am
Place: 226B Informatics East
Moderator: Yong Yeol  Ahn
Paper: Hybrid computing using a neural network with dynamic external memory
Justification: Let’s get a glimpse into whether we should worry about the super intelligence, or when we should start.

Date and time: Nov 18, 2016, 11am
Place: 226B Informatics East
Moderator: Nathaniel Rodriguez
Paper:
Justification:
Methods for ranking the importance of nodes in a network have a rich history in machine learning and across domains that analyze structured data.  Our advanced methods exhibit greatly improved performance despite being simple linear classification rules, and are even competitive with belief propagation.

Date and time: Nov 11, 2016, 11am
Place: 226B Informatics East
Moderator: Alex Gate
Paper:
Block Models and Personalized Pagerank
Justification:
Methods for ranking the importance of nodes in a network have a rich history in machine learning and across domains that analyze structured data.  Our advanced methods exhibit greatly improved performance despite being simple linear classification rules, and are even competitive with belief propagation.

Date and time: Nov 4, 2016, 11am
Place: 226B Informatics East
Moderator: SASAHARA Kazutoshi
Paper:
Quattrociocchi, W., Caldarelli, G., & Scala, A. (2014). Opinion dynamics on interacting networks: media competition and social influence. Scientific Reports, 4, 4938.
Justification: The advent of social media has changed the media landscape and our communication patterns, and previous models fail to capture opinion dynamics in such a highly-networked world. This paper proposed an interesting model of opinion dynamics on interacting networks, providing several insights.

Date and Time: Oct 28th 2016, 11 am
Place: 226B Informatics East

Moderator: Ian Wood

Paper:

Why Should I Trust You?”: Explaining the Predictions of Any Classifier
Motivation: Many new machine learning models are increasingly opaque in their predictions; even though we may understand the math behind the algorithm, we often don’t have a good understanding of why a model made a particular prediction. This paper attempts to provide better interpretations to classifiers’ predictions through a method to produce simpler, local models.

Date and Time: Oct 21th 2016, 11 am
Place: 226B Informatics East

Moderator: Nathaniel Rodriguez

Justification:Chosen because it had 4 whole votes going for it! And also because it can give us an opportunity to take a look at LSTMs –a popular machine learning tool that is commonly used in text analysis. But at the same time the principles in the paper (namely learning to learn) apply more broadly.

Date and Time: Oct 14th 2016, 11 am
Place: 226B Informatics East

Moderator: Sol Lim

Paper: Two’s company, three (or more) is a simplex Algebraic-topological tools for understanding higher-order structure in neural data

http://link.springer.com/article/10.1007/s10827-016-0608-6

Justification: Let’s see if algebraic topology can help us understand higher-order structure in neural data 🙂

Date and Time: Sep 30th, 2016, 11am
Place: 226B Informatics East
Moderator: Elise Jing
PaperFundamental structures of dynamic social networks (http://www.pnas.org/content/113/36/9977.full.pdf)
Justification: The paper meets an interest within our group on multiscale, temporal networks. In particular, I’m very interested in finding out what is a good way to characterize dynamic social networks.

Date and Time: Sep 23, 11am
Place: Info East 226B
Moderator: Pik-Mai Hui
Paper: Community Structure in Time-Dependent, Multiscale, and Multiplex Networks
Reason: We started our semester with a nice discussion on multiplex network, led by Alex. Multiplex network as a model is very applicable on real datasets, in different academic fields. A method to find communities in different scales in a multiplex network may come in handy in your career.

Date and Time: Sep 09, 11am
Place: Info East 226B
Moderator: Xiaoran Yan

Paper: Towards an integration of deep learning and neuroscience (Part 1)

Justification: Deep learning and neuroscience has developed a lot since their departure. It is time to look back together and see what can be learned from each other. The paper is also very popular in votes. Since this is such a long paper with a rich literature behind it, I intend to cover only the first 3 sections. The later half, especially section 4, involves more detailed brain functionalities that I am not qualified to take on.


Date and Time: Sep 02, 11am
Place: Info East 226B
Moderator: Alex Gate

Paper:

The physics of spreading processes in multilayer networks

Manlio De Domenico, Clara Granell, Mason A Porter & Alex Arenas

Abstract
The recent multilayer approach for modelling networked systems explicitly allows the incorporation of multiplexity and other features of realistic systems. It allows one to couple different structural relationships by encoding them in a convenient mathematical object. It also allows one to couple different dynamical processes on top of such interconnected structures. The study of multilayer networks has also revealed new physical phenomena that remain hidden when using ordinary graphs, the traditional network representation.

Date and Time: Aug 12, 11am
Place: Info East 226B
Moderator: Sol Lim

Paper:

Bennett CM, Baird AA, Miller MB, Wolford GL 15th Annual Meeting of the Organization for Human Brain Mapping. San Francisco, CA; 2009. Neural correlates of interspecies perspective taking in the post-mortem atlantic salmon: an argument for proper multiple comparisons correction.

http://users.stat.umn.edu/~corbett/classes/5303/Bennett-Salmon-2009.pdf

Eklund, Anders, Thomas E. Nichols, and Hans Knutsson. “Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates.” Proceedings of the National Academy of Sciences (2016): 201602413.

http://www.pnas.org/content/113/28/7900.full

Errata for Cluster failure: http://blogs.warwick.ac.uk/nichols/entry/errata_for_cluster/

Justification: Multiple comparisons correction, in particular, in neuroscience is still dangerously neglected, although it can be catastrophically detrimental to our understanding of the brain (or any kind of findings for that matter). The dead salmon study made a great satirical argument to show false positive activity from a dead salmon! The cluster failure study deals with  more field-specific issues but many people got interested in the study because it was covered by many major media with some over-excitement(?).


Date and Time: Aug 5, 11am
Place: Info East 226B
Moderator: YY
Paper: E. Bakshy, S. Messing, L.A. Adamic, Exposure to ideologically diverse news and opinion on Facebook, Science (2015)
Justification: Political polarization has been growing in many countries and the “filter bubbles” that technology creates have been pointed out as an important culprit. This paper examines filter bubble phenomenon in Facebook, which is where the largest amount of news articles and opinions are consumed.

Date and Time: July 29, 2016, 11am
Place: 226B Informatics East
Moderator: Kazutoshi Sasahara
Paper: James, R. G., Barnett, N., & Crutchfield, J. P. (2016). Information Flows? A Critique of Transfer Entropies. Physical Review Letters [Link] [arXiv]
Justification: Transfer entropy (TE) is often used to infer network structure from time series data such as neural spike trains and social interactions on SNS. However, the authors claim that TE fails to capture the effects of polyadic interdependencies, with two examples. This is import to keep in mind when analyzing empirical data.

Date :  July 22, 2016
Moderator : Jaehyuk Park
Justification :
1. In data-driven researches, especially in social science where controlled experiments are not much allowed, it is difficult to infer causality. This paper suggests some possible ways to infer it using machine learning techniques.
2. One of causal inference paper series in recent PNAS issue, so some parts can be also applied to outside of social science.

Date :  July 15, 2016

Moderator : Xiaoran Yan

Paper : Higher-order organization of complex networks
http://science.sciencemag.org/content/353/6295/163
Justification :
1. Interesting community structures based on higher order structures
2. A new paper with a lot of votes

Date :  April 29, 2016

Moderator : Pikmai Hui

Paper : Predicting Positive and Negative Relationships in Large Social Networks
Justification :
1. Number of vote on Google Doc
2. This is a very Machine-Learning type of paper attempting to solve a problem relevant to network science. I think it would interest people.
3. 12 pages with lots of figures + 2 pages reference. It is approaching the end of semester, and some of us may be relatively busy. This is a relatively short reading to fit the schedules.

 

Date: 4/22/16

Moderator: Azadeh

Paper: Global Diffusion via Cascading Invitations: Structure, Growth, and Homophily

Linkhttps://www-cs.stanford.edu/people/jure/pubs/linkedin-www15.pdf

Justification: This  paper analyzes the effect of homophily in global diffusion. I am curious to learn more about their approach!


Date: 4-15-16

Moderator: Ian Wood

Paper: Time Complexity of Evolutionary Algorithms for Combinatorial Optimization

Link: http://link.springer.com/article/10.1007/s11633-007-0281-3

Justification: Evolutionary algorithms are a heuristic search method for exploring complex parameter spaces. They are introduced early in the complex systems program, and are a good example of a bio-inspired method for machine learning. However, it is unclear how good the fit or efficient the search might be for a particular problem. This paper describes some of the methods to apply bounds to describe the run-time complexity of such algorithms applied to a number of different problems, along with some discussion of when to use variations.


Date: April 8, 2016

Moderator: Yy

Paper Title: Mastering the game of Go with deep neural networks and tree search

URL: http://www.nature.com/nature/journal/v529/n7587/full/nature16961.html

Justification:  Human vs Machine, with YY as the moderator who knows both Go and deep learning. What else do you need?


 

Date: April 1, 2016

Moderator: Xiaoran

Paper Title: Deep learning

URL: http://www.nature.com/nature/journal/v521/n7553/pdf/nature14539.pdf

Justification:   The alphago game really got me interested in the theory of deep learning. This should be helpful to prepare for the alphago paper when YY present it the week after the next. Many of you know a lot more about deep learning than I do, I look forward to your opinions.


Date: March, 11, 2016

Moderator: Pikmai Hui

Paper Title: Tracking the Evolution of Communities in Dynamic Social Networks

URL: https://csiweb.ucd.ie/files/ucd-csi-2011-06.pdf

Justification:   Recently we have acquired access to the Web-of-Science dataset. While Rion is working on structuring the data, I think it may be of our interests to think of ways to keep track of series of partitions/communities. Many studies focuses on unique snapshot of a network, and assume implicitly that the network remain static within the period it is collected. With datasets that span a long time period, such as the Web-of-Science, it is possible to construct ordered sequence of snapshots and study the evolution of communities across snapshots. This paper describes one way of doing it.


Date: March 4, 2016

Moderator: Yizhi Jing

Paper Title: Causal Entropic Forces

URL: http://www.alexwg.org/publications/PhysRevLett_110-168702.pdf

Justification:  One of the highly voted papers. This paper brings together interesting topics in information theory, causality, intellegence and evolution by including them into a thermodynamic model. Many of us are familiar with some of these topics, but it may be interesting to look at them from a novel perspective.


Date: Feb 26, 2016

Moderator: Ian Wood

Paper Title: Latent Dirichlet Allocation

URL: http://dl.acm.org/citation.cfm?id=944937

Justification: This paper presents the original Latent Dirichlet Allocation model. This is an important method for topic modeling, and the paper is also a good example of Bayesian modeling. I have read papers using and extending the model, but I think many of us have not read the original paper, so I think it would be informative to step back and discuss this paper.


Date: Feb 12, 2016

Moderator: Jaehyuk Park

Paper Title: Scaling and universality in urban economic diversification

URL: http://rsif.royalsocietypublishing.org/content/13/114/20150937

Justification: This paper presents a simple mathematical derivation of the universality, and provide a model, together with its economic implications of open-ended diversity created by urbanization, for understanding the observed empirical distribution. I chose it because 1) it would be helpful to many of our members, who are interested in urban economic mobility & development, and 2) this seems also good to be linked with researches done by Rossano (who visited our school in the last week), to extend our discussion.


Date: Feb 5, 2016

Moderator: Azadeh Nemtzadeh

Paper Title: Determinants of Meme Popularity

URL: http://arxiv.org/pdf/1501.05956

Justification: This paper introduces a generative model for information spreading in online social media, focusing on two factors affecting meme popularity. These two factors are the structure of social network and the memory time of users.


Date: Nov 29, 2016

Moderator: Xiaoran Yan

Paper Title: Structure and inference in annotated networks

URL: http://arxiv.org/abs/1507.04001

Justification: This paper generalized stochastic block model by incorporate meta data on nodes. Using a Bayesian framework, the community detection algorithm is able to leverage the annotations if they are correlated with community structures, or ignore them when they are independent.


Date: Nov 22, 2015

Moderator: Krishna Bathina

Paper Title: Principles of assembly reveal a periodic table of protein complexes

URL: http://science.sciencemag.org/content/350/6266/aaa2245.full

Justification: This paper examines patterns in the quaternary structure of human proteins by looking at three independent sets of data. These patterns are organized into a ‘periodic table’ and are used to predict currently undiscovered structures. I chose this paper because it is a good example of analyzing, predicting, and validation on a data set.


Date: Nov 20, 2015

Moderator: Jaehyuk

Paper Title: Secondhand Brokerage: Evidence on the Importance of Local Structure for Managers, Bankers,and Analysts.

URL: http://www.jstor.org/stable/20159844?seq=1#page_scan_tab_contents

Justification: This paper examines under what specific topological situation the structural hole position can have advantages throughout the dataset of bankers. I chose this because it can provide us a sociological approach to network analysis that rarely be considered in our field.


Date: Nov 6, 2015 Moderator: Azadeh

Paper Title: Finding Statistically Significant Communities in Networks

URL: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0018961

Justification: We had read several papers related to network community. I find this paper pretty relevant to some of the projects that the reading group attendees are involved with. In this paper, authors proposed a method to detect clusters in networks accounting for edge directions, edge weights, overlapping communities.


Date:  10/30

Moderator:  Thomas

Paper title:  Quantifying Creativity in Art Networks

URL: http://arxiv.org/abs/1506.00711

Justification:  Interesting article on computational creativity. Their method of quantifying creativity via a creativity implication network could be applied towards any cultural product.


Date: Oct 23, 2015

Moderator: Yizhi

Paper Title: Memory in network flows and its effects on spreading dynamics and community detection

URL: http://www.nature.com/ncomms/2014/140811/ncomms5630/pdf/ncomms5630.pdf

Justification: One of the most voted articles. This paper argues for the importance of taking memory into account in modeling spreading process and doing community detection. This was made by studying the effects of second-order Markov dynamics in real-world networks.


Date: Oct 16, 2015

Moderator: Qing

Paper Title: Influence maximization in complex networks through optimal percolation

URL: http://www.nature.com/nature/journal/v524/n7563/full/nature14604.html

Justification: Most voted article. The authors proposed a method that maps the problem of identifying influencers onto optimal percolation, and found that a large number of optimal influencers are low-degree nodes.


Date: 10/9

Moderator: YY

Paper title: word2vec

URL: http://yyahnwiki.appspot.com/word2vec

Justification: word2vec is a very cool ‘deep-learning’ technique to learn dense vectors that represent words, from a large set of documents. It allows analogical inferences (e.g. v(‘king’) – v(‘man’) + v(‘woman’) ~ v(‘queen’) and people demonstrated that the word vectors and document vectors can improve various text mining tasks (e.g. sentiment analysis) significantly.


Date: September 25, 2015

Moderator: Pikmai

Paper Title: Topological data analysis of contagion maps for examining spreading process in networks

URL: http://www.nature.com/ncomms/2015/150721/ncomms8723/full/ncomms8723.html

Justification: Modern networks, digital or physical, often have “long-range” edges, which connects nodes that are otherwise far away from each other. They can be flights from Asia to America, or likings of a video from a different culture. We all know this is called the small-world effect. This paper considers this effect as noises to a classical concept of geometric contagion. Yes, it builds new models. More importantly, it shares a new angle from which we can construct a low-dimensional structure of networks. This is quite useful in our research in general.


Date: September 18, 2015

Moderator : Ian

Paper Title:  Mapping Change in Large Networks

URL: http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0008694

Justification:


Date: September 11, 2015

Moderator: Nathaniel

Paper Title: Symbolic regression of generative network models

URL: http://www.nature.com/articles/srep06284

Justification: In the past we have covered a couple different network generating models. In continuing that theme I wanted to try a paper that takes a rather different approach toward generating networks. In previous papers we saw hyperbolic geometry and the block model used in combination with bayesian methods to find generative models for networks. This paper uses evolution to train a symbolic representation of the network. We can go over what advantages and disadvantages that this approach has relative to the other methods.