Sign up and get paid for class action lawsuits easily

If there is an industry rife with high payment fees and cooperative problems, it is the class action lawsuit business. With around $6B in payouts each year in the US, it’s a lucrative, early market for smart contracts to emerge.

Technology: A blockchain allows for a self executing contract. This in turn can automate payouts based on the proportion of harm. For example, if an energy drink company were to sell its drink with 40mg of caffeine versus 80mg, a purchaser of that energy drink would be eligible to compensation, which would be determined by the number of energy drinks purchased (up to a limit). A smart contract could be coded to automated this payout system.

Why blockchain is better: Most payments are made through the mail which has a high cost. Payments under $.10 are often not even fulfilled due in many cases. People often move, which stops checks from being cashed. In one case, nearly ⅓ of the funds from a $100M lawsuit were unclaimed.

Value Prop: Blockchain can decrease transaction costs of current payment systems while creating cryptographic proof that the payments were made. It can automate the distribution of compensation based off harm.

Why it’s important: Smart contracts will be the basis of future agreements and will slowly infiltrate nations’ legal system. Startups that build a trusted solution in the courts could quickly gain validity and traction.

Introduced by Brian Bordley , brian (AT) skydeck.vc

HUMM: Brain Computer Interface Project

Ever wished you could upgrade your brain like you update an app on your phone? HUMM is building a hardware/software platform for anyone who depends on their brain for success. The first users are “cognitive athletes” in the professional esports gaming space, harnessing cutting edge neuroscience to unlock improved mental processing speeds and accelerating training. To achieve this, we are building a model for cognitive performance enhancement for competitive esports players. The data set is a combination of live game-play statistics derived from the API of League of Legends (the world’s largest esports game) and biometric data from a combination of sensors including heart rate (and derivatives such as heart rate variability), eye tracking, EEG (brainwaves) and facial expression analysis.

If you’re interested in learning data science while also helping build the future of brain-computer interface, and perhaps playing a few games along the way, then this is the project for you! We can be contacted at thinkbetter@humm.tech, or check out our esports site at www.humm.gg if you just want to stay in touch with developments.

Proposed by Iain McIntyre, Co-founder HUMM

Concept: Emotional Instability Predictive System

Introduced by Kenneth Lee,Amy Tsang, Christopher Francisco, Ludwig, The Hong Kong Federation of Youth Groups

Contact: kenneth.lee@hkfyg.org.hk

 

Problem to solve:

Emotional downturns, instability and depression could cause mental illness, or suicide in extreme case.

  • Early warning signs of can be found on online platforms (FB, Instagram, WeChat …)
  • The earlier the counsellor can identify people in need, the better he can provide support.

Approach

Past real-case conversation transcript (w/o personal identifier), users’ demographic data, and issue categories as input for machine learning. All this can be used to develop a classifier that predicts if the person has an emotional downturn.

Benefit

A service supervisor can be alerted if the person who is likely to have an emotional downturn is engaged with any counselors → Better resource management and oversee the online support network.

Read more about the in this PDF:

Youth Suicide Prevention (Emotional Downturn Detector)

Concept: AI Text Customizer

Introduced by Elaine Wong, Thomas Poon, John Kwok, Stewart Chu and Rocky Shek

Contact: elainew@hkbu.edu.hk

About the Project:
Our partners in Hong Kong are proposing a project in which deep learning techniques together with NLP approaches are applied to the problem of linguistic style transfer for text. For example, style transfer in the domain of natural language can be applied to customize a text so that it is suitable to publish on different social media platforms, hence relieving the author of the tedious work of tweaking the text for different audiences.
Read more in the Project PDF created by experts from Hong Kong here: AI Text Editor_v2

Concept: Automated asymmetry detection in digital mammography using u-net deep learning network

Introduced by Ensi Khalili Pooya, n30.khalili@gmail.com and Ashkan Yousefi, PhD, ashkan.yousefi@berkeley.edu.

Introduction:

Population-based breast cancer screening programs with mammography have proven to reduce mortality and morbidity associated with advanced stage of disease.

Screening mammography included two views of each breast (craniocaudal-CC and mediolateral oblique-MLO). One of the suspicious mammographic findings for mass is asymmetries, findings that represent unilateral deposition of fibroglandulair tissue not confirming completely to definition of a mass. Asymmetries are classified in four groups: 1. Asymmetry : as an area of fibroglandulair tissue visible on only one mammographic projection, mostly caused by superimposition of normal breast tissue.2. Focal Asymmetry: visible on two projections, hence a real find in rather than superposition.3. Global Asymmetry: consisting of an asymmetry over at least one-quarter of breast and is usually a normal variant.4. Developing Asymmetry: new, larger and more conspicuous than on a previous examination.

Among these four types we want to work on type 2(Focal Asymmetry) because this type has to be differentiated from the mass. For this reason radiologists request additional view, Focal Compression Magnification view (FCMV), from the desired Focal Asymmetry to see if there is a real mass under it(asymmetry does not resolve in FCMV) or not . Each additional view has radiation dose equal to one chest X-ray.

In this study we aim to produce AI algorithm to detect Focal Asymmetries, with accuracy similar to an expert radiologist, which can predict the probability of being resolved in FCMV (low/high probability). This computer-aided detection system help radiologist to evaluate mammographic exams faster and also reduce the number of requests for FCMV in order to reduce radiation exposure dose to patients.

Patient selection :

This study will conduct with anonymized data retrospectively collected from our institutional archive.

Digital mammogram exams from women attending the screening program at our hospital for diagnostic purposes from 2016 to present.

All exams are bilateral and included two views (craniocaudal CC and mediolateral oblique MLO).

The images were acquired by HOLOGIC, USA.

Conclusion:

In this abstract we presented computer-aided detection and decision system which can detect Focal Asymmetries in digital mammography and can predict which one has low probability to resolve in FCMV so radiologists must take additional projection from them and which one has high probability and radiologists can overlook.

This strategy reduces exposure radiation dose to patients and also reduce the number of patients recall to a hospital and reduce costs.

 

Concept: Learning AI / ML by Modeling Human Social Trends as a Time Series

Introduced by Milind Gadre, Senior Director R&D at VMware, Inc.

Human society today is being roiled by a combination of negative / disruptive trends such as the rise of authoritarianism and populism, fueled by xenophobia and a fear of social change. At the same time, we can see a rise in positive / cohesive concerns for expanded human rights such as LGBT rights and empowerment of women. Religious fundamentalism continues to drive many societies and contributes to several of the negative / disruptive trends identified above. At times, it feels like we are regressing to social behaviors of the past.

If one looks at humanity in the aggregate, it appears that variations of the same social trends rise and fall in waves across seemingly disconnected societies and time frames. There is an ebb and flow to these trends … positive trends seem to almost inevitably give rise to negative ones and (sometimes) vice versa. Trends may start as one or many localized events of limited scope that fade without effect, but sometimes coalesce and balloon to global scope (see for example – the Syrian Civil War started with a little graffiti). Furthermore, communications today is near instantaneous at a global scale, and it amplifies or dampens the growth of a trend.

Questions

Have you ever wondered about the possible evolution of human society? Is a continuous oscillation between cohesive and disruptive trends the “normative” state for human societies? Will the current set of negative trends be amplified leading to conflict on a global scale (as happened in the early parts of the 20th century)? Will there be a new cold war between authoritarian and liberal societies? Or will the positive / cohesive trends prevail, leading to a better society for all? How will external disruptions such as climate change affect the outcomes?

Is it possible for us to learn from the past when similar trends prevailed? How can we analyze past social data to make predictions of how human society will evolve?

Ideas from Data Center Ops Management

Most large-scale data analytics is now stream based. Time series analytics is a common approach to data center ops monitoring applications (e.g. Log Insight or Splunk for log analytics, Wavefront or Datadog for time series-based metrics). Commercial data center Operations Management applications routinely monitor, analyze and provide actionable feedback on highly complex, seemingly random data, at scale. Ops Management solutions can auto-identify a “normative state” and “violations” for data center systems simply by evaluating operating data as a time series over some period of time.

Hypothesis

Today, we have a large enough population size (over 8B humans), instant global communication – and the tools to analyze data efficiently at scale. By modeling human social evolution as a stream of events – a time series – we can apply AI / ML techniques to correlate events and to extract (even possibly predict) social trends.

What It Is Not

This approach is not the same as micro-targeting – which is employed by advertisers, political parties and the like (e.g. Cambridge Analytica). Hence, we do not need personal / private information. The proposal works by analyzing events in the aggregate as a time series, not on the basis of individual personal characteristics.

Challenges / Learning Opportunities

The challenges fall into 3 buckets:

Challenge 1 – Collecting Data

We need access to both historical and real-time event data.

Event: Any happening such as demonstrations, political events, court cases, etc.

We have to concern ourselves with both

  1. Historical event data –

a. We need access to historical data

b. We would need to crowdsource the shaping of the data as an event stream

Contemporary event data – this would rely on

a. Mining of data streams such as Twitter feeds

b. Mining of news sources such as Reuters and AP feeds

c. Social media apps for crowdsourced submission of real-time events

Challenge 2 – Shaping the Data

The event data will need to be shaped as a time series and tagged with a variety of information so as to support the AI / ML routines.

Some of the possible ways in which the event data would be tagged –

Basic Tags on the Event: date, time location, country, number of people involved, etc.

Social Trend Tags on the Event: Racism, xenophobia, authoritarianism, populism, civil rights, etc.

Information relation to the Event locale (incomplete list): Coefficient of social friction (0 to 1), economic conditions, Gini coefficient, weather, climate change, etc.

Challenge 3 – Developing the AI / ML Analytics

The final challenge is the actual development of the AI / ML analytics and the fine tuning so that we can

1. Based on historical event data, accurately model and predict historical trends – this is the Learning part.

2. Predict future trends based on a combination of historical and contemporary event data.

In Summary

This project offers a challenging opportunity for students that are interested in learning about AI

/ ML and are also interested and curious in making sense of the complex political and social trends in our societies. There are opportunities to learn about the end to end challenges involved in developing a real-world AI / ML application – from collecting the data, then shaping it, followed by developing and fine-tuning the algorithms. There are challenges to be addressed in scraping social media and building apps for collecting contemporaneous events. There are opportunities for developing interesting data visualizations and user interfaces. Finally – there are opportunities to write and publish technical papers.

Concept: Deep-learning startup focused on optimizing supply chains for manufacturers

Introduced by Moataz Rashad, Founder & CEO of DeepVu, http://deepvu.co, @deepvuhq, @moatazr.

DeepVu is a deep-learning startup focused on optimizing supply chains for manufacturers. We work with tier-1 manufactures in the US and Asia.

One of the secondary sub-use-cases that we encounter involves forecasting the price of certain commodities that are key constituents of our customer’s products Bill of Materials. If it is often needed to forecast the price of that commodity (for example, copper, PVC, aluminum, IronOre etc) several months into the future and in some cases a year out in order to inform our model’s predictions of the price of the manufactured parts.

For this competition project, you’ll be given a commodity’s market data, typically 5 years, and you’ll have the following attributes (open-price, close-price, trading volume, day high, day low, etc.)
You do have the freedom to add additional columns that you think may help enrich the intelligence of your model, for example, the price of gasoline, GDP data, etc.
You have the freedom to choose any deep-learning or traditional machine-learning model of your choosing.

Framework Requirements:
Only Tensorflow or PyTorch are allowed and we encourage you to use GPUs.

Testing:
We recommend you split the data-set 80% for training and 10% validation and 10% testing

Performance Goal: Mean Absolute Error of 7.5% as measured against actuals prices for
time period 4 weeks out (from the last date in the training set given) to 12 weeks out.

Submission Deadline: April 15th, 2018. You’ll need to submit:
a. your source code in python (GitHub repo is fine)
b. README file with steps that include any framework/library dependencies and how to train and run inference.
c. a csv file with the predicted vs. actual on given test-set
d. csv file with predictions for daily prices 4 weeks out till 12 weeks out.
e. Performance numbers and analysis document

Award:
1st place: $2500 cash award + priority placement for a paid summer internship
2nd place: $1200 cash award + priority placement for a paid summer internship

Concept: 3D body scanner to revolutionize how we discover & shop for clothing

Introduced by Richard D. Berwick, Co-founder, Twindom, 3D Scanning, 3D Printing, and 3D SAAS, (765) BER-WICK | richard@twindom.com
What Twindom does:
Twindom is leveraging the photorealistic quality of our 3D body scanner and the data we’ve captured to date to produce Drapr, a Virtual Fitting product. Our goal is to revolutionize how we discover and shop for clothing by applying bleeding edge research in body scanning, body modeling, physics simulation, geometry processing, machine learning and more.
Who we are:
As a team, we (1) seek to be world class, (2) get things done, (3) love to learn, (4) thirst to build products customers love, and (5) believe you have to act differently and creatively to stand out. If you believe in these principles, you’ll fit in well at Twindom. We’re looking for the best in the world – and if that is you, you’re a good candidate for the job.
What you’ll do:
You will be responsible for designing and developing major components of the virtual fitting pipeline. Our small and dedicated team is currently tackling a wide variety of problems in computer vision, including image segmentation, mesh segmentation, image recognition, mesh manipulation, rendering, body modeling, physics simulation, and 3D scanning. The challenges we encounter are endless — and as a current/past Cal student, you know how to solve problems.

Concept: Science-based sleep coaching

Introduced by Dax Vivid, Postdoctoral Researcher at UC Berkeley, daxvivid@berkeley.edu, and keyser@berkeley.edu.

About SleepWeb:
We are harvesting the capabilities of existing sensors to provide science-based sleep coaching. Our team includes a mechanical engineer and a biologist. We are looking for software developers to join our team. This is an opportunity for specialists in AI, statistics, and machine learning to develop the backend of software that integrates basic environmental and physiological constituents of good sleep then offer a framework for advice based on user inputs.

Concept: Creating a real-time newsbot

Introduced by Jason Best, Vectr Ventures, jason.best@gmail.com, David Law, SCET Berkeley, dlaw@berkeley.edu, and Anand Gomes, CEO & Founder at RiskEx (http://www.riskex.io/, https://www.linkedin.com/in/anandgomes).

Concept: Creating a real-time newsbot that scrapes and publishes the most relevant news based on one or more tickerized asset classes and related filters. These news items are then inputted to a single-page web-widget. This will be the first truly “unbundled” news bot for the financial markets.  

Problem: Even in 2018, there is currently no free or inexpensive way of aggregating news that is relevant to a specific asset class in a single-page-application by simply toggling the asset class, using its ticker. Certain paid services such as Bloomberg and Reuters do a pretty good job but they cost $25K / year and there is no way to “unbundle” their news offerings. This is similar to what is happening with cable television and the desire to purchase only the specific channels that you watch/care about.

To learn more, please click HERE.