CMSC 798E Graduate Seminar in Computer Science

posted in: Fall2013, Graduate | 0

This is a 1 credit seminar. We hand in summary of lectures to Prof. Jeff Foster every week.

CMSC 798E Ruofei Du’s Summary for 09/13’s lecture

* What was the most interesting thing you learned during this talk?

In the lecture on Sep. 13th – “Duking It Out at the Smartphone Mobile Mapping App Corral: Apple, Google, and the Competition”, Prof. Hana Samet gave us a informative talk on the problems discovered in industrial state-of-the-art map apps.

The most interesting thing I learnt from the talk is the importance of consistency (when designing a map App but not limited in map Apps). Beforehand, when I encounter consistency problems when exploring on a map App (e.g. the label disappeared), I did not pay much attention to the problem but just keep exploring until I see the labels clearly. Nevertheless, the overlook of consistency led to not only bad user experience but a waste of time, confusion of users, distrust of the routing algorithms.With bad consistency in the maps’ APIs, reading newspaper, photos and videos could be a terrible experience for users.

* How does the material presented in the talk relate to your research interests?
I am working on a Google Street View project with Prof. Froehlich. And I am interested in Human-Computer Interaction and Computer Vision. This talk provides me good vision of how much consistency values when designing geometric systems. Consistency should be kept everywhere: panning, zooming, overlapping, sibling, wraparound… Only in this way, can we provide friendly and effective algorithms and systems to the users.

CMSC 798E Ruofei Du’s Summary for 09/20’s lecture

* What was the most interesting thing you learned during this talk?
Prof. William Gasarch gave a impressive talk about “The Distinct Volumes Problem” with full of passion on Sep. 20. The most interesting thing I learnt from today’s talk is his dimension-reduction method employed in his research work. First of all, he proposed 2 examples about rainbow sets (a set of points where all of the distances are distinct) on R^2. Afterwards, we investigate into the 1-dimension case, which greatly simplifies the hard problem. After solving the 2-dimension case, Dr. Gasarch proceeds to general d case. Then he seeks help from a high school student Sam. This is a useful tip for research: Seek and get help from others when stuck. The inference made from low dimension to high dimension is indeed useful. Amazingly, they finally proceed to infinite rainbow subsets of size \alpha. His wife inspired him to work on getting the idea from AD. And Dr. Gasarch got help from Douglas Ulrich.

 

   * How does the material presented in the talk relate to your research interests?
Most points have been mentioned on the paragraph above. A little summary:
1. Start from easier case and try to make inference on more difficult ones.
2. Use dimension reduction technique or add limitations could be a help. Try to look into other things when stuck.
3. Get help from others, collaboration makes perfect.
4. Communicate as much as possible with people around you. Get inspiration!
5. Give talks on them could get help! Make things happen!
6. You must be broader, and then focus. 

   And I firmly believe these tips could help to my future research.
CMSC 798E Ruofei Du’s Summary for 09/27’s lecture
  * What was the most interesting thing you learned during this talk?
Prof. Ashok Agrawala gave us a lecture on “Content-Aware Computing”. I like his idea on automatic tailoring of information and services based on the current location of the users. To investigate location, his team developed Horus WLAN location determination system where the intensity of the signal from Wifi access point is measured to calculate the location. The accuracy of less than 2ft in A.V. Williams is amazing! Additionally, Prof. Agrawala also presented his demo of M-Urgency at UMD. This next-generation safety alert system introduced context-aware data by recording real-time videos and user locations. In a word, finding the relationship between context-aware data and our desired information is interesting. Nevertheless, to build a real-time system and solve the problem regarding context-aware information is challenging due to management of spatial data, security issues, scalability…* How does the material presented in the talk relate to your research interests?
Though I am not that into network and system research. His projects greatly inspire me to think of the context-aware information when designing new systems in HCI: indoor / outdoor; the challenge when involving location;  Finally, considering the rate of information production is accelerating, to collect Twitter data together with geo-location is interesting. The ultimate goal of context-aware computing is to mprove the quality of life!

CMSC 798E Ruofei Du’s Summary for 10/04’s lectures
 * What was the most interesting thing you learned during this talk?In today’s talk, five professors offered five impressive talks, in which I love Prof. Findlater‘s best. Her observation in accessibility problems in daily life led to her many splendid projects: Beyond QWERTY, Personalized Input, Motor impaired touchscreen use, Age-related Differences in Input…. Other notes are listed below:

1. Prof. Marshini Chetty introduced HCI and Networking to us. The interesting part is that she grasped the demand for users on high speed Internet: Whether the devices are connected correctly, what’s the major traffic cost in the home network? How is the bandwidth?

2. Prof. Dana Dachman-Soled introduced her research in cryptography to us. She used a lot of vivid scenarios to illustrate the motivation of cryptography against physical attacks and secure multiparty computation.

3. Prof. Tudor Dumitras introduced his research in Improving System Security with Big Data Techniques. His system WINE aims at big data security analysis. The challenges in his security research: Identify and measure zero-day attacks (Find the needle in the haystack) & ensure generally applicable and repeatable results…

5. Prof. Charalampos (Babis) Papamanthou give a presentation on “do you trust your cloud”. I can’t believe at first glance that Hotmail, Amazon, Gamil, Dropbox, Microsoft are facing data corrupting issues on their cloud. So reliability is indeed a problem. Privacy-preserving should always be considered when designing a system.

  * How does the material presented in the talk relate to your research interests?    

I’m already working at the HCIL with Prof. Jon Froehlich. So Prof. Findlater‘s research greatly relate to my research interests 🙂 My ultimate goal “Design the future” matches very well with Prof. Froehlich and Prof. Findlater. I’m doing the HandSight project. The general idea is to put nano cameras onto fingertips / gloves so that the blind people can “see” through their hands.
CMSC 798E Ruofei Du’s Summary for 10/11’s lecture
* What was the most interesting thing you learned during this talk?
In today’s talk. Prof. Elaine Shi gave us an introductory and interactive talk on How to Store your Data Safely in the Cloud. Storage outsourcing introduced problem such as corrupted data and privacy concerns. Specifically, we care about ensuring integrity (authenticity + freshness) in outsourced storage.
The challenge is that storing hashing data might be too expensive and inefficient.
1) Hashing a block is a good idea.
2) Merkle Hash Tree greatly improves the efficiency when only few of the node files are updated as well as providing dynamic updates.
How shall we check the integrity?
1) Check all: O(N)
2) Check k randomly: Prob = (1-t/n)^k -> 1 – O(1) when t->1
3) Again use the idea of Merkle Hash Tree!
Finally, Dr. Shi introduced her work on dynamic proofs of retrievability: the high level idea is to maintain a hierarchical log; from the leaf to the top method.

 
    One lesson I learnt from this talk is to transfer one idea in one field to another one and investigate the domain knowledge, which could solve another open problem!* How does the material presented in the talk relate to your research interests?
Though I’m not doing research in cloud computing, the issues brought by cloud computing is pervasive in everyone’s daily life. Amazingly, the insights behind the beauty of data structure successfully solved the data integrity problem in cloud computing.

CMSC 798E Ruofei Du’s Summary for 10/18’s lecture
* What was the most interesting thing you learned during this talk?
In Prof. Dana Nau’s lecture “What a Long Strange Trip It’s Been” today, he talked about his own experience on the way pursuing his Ph.D. degree. The most interesting I learnt from this talk is that when he was researching upon game theory for several years, there seemed to be little hope of breakthrough. However, it is his perseverance and his advisor’s encouragement that finally leads him to a great success.

* How does the material presented in the talk relate to your research interests?
The beauty of game theory and artificial intelligence really fascinate me. Moreover, Prof. Nau’s life story teaches us a good lesson: to concentrate our energy, thought and capital exclusively upon our engaged research area and do not give it up easily is the vital key to success in academia.

CMSC 798E Ruofei Du’s Summary for 10/25’s lecture
* What was the most interesting thing you learned during this talk?
 
In today’s lecture, my advisor Prof. Froehlich gave us an impressive talk on “Making with a Social Purpose”. I am really interested in his passionate research path towards his Ph.D. thesis “HydroSense”. Generally speaking, the motivation of HydroSense is “Designing4Good” and encourage every family to save water. The water-pressure and machine-learning based sensing system is so clever that it can infer fixture-level usage for the entire home from a single point with an overall accuracy of 90%. Moreover, he explored further to visualize the data by interactive Aquatic Ecosystem View, Rainflow View, Time-Series View, which greatly inspire me for my Information Visualization course project 🙂 Finally, he encouraged us to be independent, creative, good builders, self-disciplined, ambitious, fearless, passionate, robust, which I might benefit from in the following years during my graduate studies and research.
 
  * How does the material presented in the talk relate to your research interests?

Actually, I am conducting research on HandSight and helping on Kotaro’s SideWalk project mentioned in today’s talk. Prof. Froehlich also brings up the accessibility-aware navigation systems, which might a potential research paper in the future for SideWalk project. All the others projects mentioned today gave me a broader view of what is going on at the Makeability Lab and how they manage to contribute ideas to the physical world. Finally, I would also learn the way how Prof. Froehlich effectively and passionately convey the presentation.
Here’s CMSC 798E Ruofei Du’s Summary for 11/08’s lecture.

* What was the most interesting thing you learned during this talk?

In today’s lecture, Prof. Adam Porter gave us an introduction on “Testing Large-Scale, Configurable Software Systems”. Prof. Porter led us explore the problems brought by potential 1,000,000,000,000+ configurations and corresponding various execution paths in software. That’s really interesting since I have never thought of combining configurations together for testing more efficiently and how to combine configurations to guarantee the coverage of testing all the code? Initially, the path tree algorithm took them 2 weeks on 40 GPUs to analyze the code. To make it feasible, they finally proposed iTree (Interaction Tree) to generate a small configuration sample ,instrument the system, and observe its behavior. Machine Learning techniques was used to discover potential interactions. The time and efficiency was greatly improved. With same numbers of configurations, iTree coveraged more lines of code and incurred only 0.5% CPU over-headed.

 
Research Path deserve learning:
perform additional empirical studies
learning initial configuration options
applying iTree to test cases in addition to configurations.
* How does the material presented in the talk relate to your research interests?

On the idea side, though I have developed software with hundreds of configuration files before, I have rarely thought of testing certain configuration settings efficiently. Nevertheless, Prof. Porter’s lecture reminds me of finding the “hall of shame” from everywhere in life. After great efforts on the new idea, we could possibly convert “hall of shame” to “hall of fame”.
Additionally, the techniques on testing configuration proposed by Prof. Porter make me review the configuration parts of my research projects. To iteratively develop a real-time research demo, there are tons of configurations file to deal with. However, I paid little attention on testing a set of configuration settings.

CMSC 798E Ruofei Du’s Summary for 11/15’s lecture
* What was the most interesting thing you learned during this talk?
 
In today’s lecture, Prof. Héctor Corrada Bravo introduced his latest research on “MethyIFlow, Methylation haplotype reconstruction from bisulfite-converted DNA sequencing”. It’s really hard to understand the contents without prior knowledge on computational biology. Basically, they are curious how is the letters of DNA sequence read and parsed? We need grammar! The most interesting thing I learnt from this talk is that how machine learning and network flow algorithms are applied in a practical computational biology problem. For example, given a set of mapped reads, how to determine composition of methylation fragments? The composition includes fragment identities (metalization patterns present) and fragment abundances (abundance of each pattern present). And how many unique patterns; community diversity. They finally figure out some patterns from the fragments and reconstruct haplotype by network flow algorithms. 
 
  * How does the material presented in the talk relate to your research interests?
I have no prior experience in computational biology. But I learnt a log of open questions from today’s talk. e.g. How are cell populations changing in normal and tumor? And these critical questions are greatly concerned with human beings’ health. Another interesting point is how algorithms and models are built in practical problems. For example, how network flows are used in haplogyping. Prof. Bravo aligns reads as edges, consistent overlap as nodes, read coverage as min capacities and solve the problem by maximum flow.
CMSC 798E Ruofei Du’s Summary for 11/22’s lecture
* What was the most interesting thing you learned during this talk?

In today’s lecture, Prof. Shneiderman and Megan Monroe gave us a splendid talk on “Interactive Visual Discovery in Temporal Event Sequences: electronic Health Records & Other Applications”. The most interesting thing I learnt from this talk is that how patients and doctors get significant benefits from the health visualization such us “LifeLines2” & “LifeFlow”->”EventFlow”. The temporal health data can be amazingly categorized and rendered in a single graph, which helps doctors and researchers find the patterns in the health data in an intuitive way. The basic idea of LifeFlow is to categorize different events based on temporal length. This idea to then extended to EventFlow, which can visualize temporal event records and explore various patterns. Megan Monroe also gave awesome analytics on UMD Basketball match data.

* How does the material presented in the talk relate to your research interests?

Information visualization is an indispensable topic in Human Computer Interaction field. Generally speaking, I believe every CS researcher needs good information visualization skills in figures of papers. If the visualization is clear and novel, it makes more sense to the reviewers and audience, which could in return bring more acceptance rate and citations 🙂 In the future, I might have to deal with a lot of data, hopefully the data could be visualized in a intuitive way inspired by this talk.

CMSC 798E Ruofei Du’s Summary for 12/06’s lecture
* What was the most interesting thing you learned during this talk?

In today’s lecture, Prof. Zia Khan talked about his research on fluorescence microscopy. I never thought about observing, nor segmenting and tracking dynamic behaviors of cells and tissues before. Amazingly, the research work led by Prof. Zia Khan provided a 3D cell outline segmentation algorithm. His EDGE4E tool enables quantification of the dynamics of cell shape changes, cell interfaces, and neighbor relations, at single-cell resolution during complex morphogenetic events. It’s really a success story combining computer vision and biology.

* How does the material presented in the talk relate to your research interests?

From the imaging data, Prof. Zia Khan utilized computer vision to measure into quantitative models and new hypotheses. Most CVPR papers I read before care more about the efficiency and accuracy of a certain algorithm, rather than what specified research problem could it done in another field. Though I am not into computational biology, I am glad to see Prof. Zia Khan’s success in designing while transferring state-of-the-art algorithm to high-quality tracking and visualization of fluorescence microscopy. I should definitely learn from him. 

In addition, I was amazed to Prof. Zia Khan’s splendid work on CVPR and ECCV on tracking. Then his transferred his awesome work and get published on Nature and Science. His successfully story is indeed a great example for us first-year graduate students..

Leave a Reply