The following is my final presentation for my Design for Climate Change class taught by Peder Anker. Additionally, it was selected to be presented at New York University’s 2040 Now exhibition, an event dedicated to ideas of how NYU can become carbon neutral by 2040. To see the whole deck [click here](https://pitch.com/v/washington-square-park-redesign-jmnm83).

Our group presented a redesign of NYU’s entire Washington Square campus. In the following slide deck, we describe how the new campus can not only become sustainable and climate friendly through the reduction of automobile emissions and introduction of green space, but we can also build a stronger community for those living in the surrounding areas, by creating new spaces for people to gather and local merchants to sell goods.

The deck begins with an identification and description of the problem: The streets around WSP and NYU's campus don't adequately serve the needs of the community. We then lay out our inspirations and design references, followed by on-sight research conducted to find the real needs of the community. We also include research into the materials we plan to use. Finally we present our solution to how the area can be designed to better serve those who use it.

Our solution plans to bring back hope to the community, by improving their quality of life with designs that create a healthier environment, and raise the mental and physical health of community members.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FUKLkhMFOJFbDx4g19Xqd%2F428ff7d2-62e4-4428-933a-2a0267a3345f.png?alt=media&token=b3032864-f348-43ff-b088-c99882404ee8","id":"428ff7d2-62e4-4428-933a-2a0267a3345f","width":2354,"height":1220,"filename":"Screenshot 2024-03-31 at 6.27.00 PM.png","type":"image/png","caption":"","border":false},{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FUKLkhMFOJFbDx4g19Xqd%2F584633d4-7a8e-4c40-aa69-7a4851433c05.png?alt=media&token=a6bcbbdc-f9cc-4b9e-ab0d-da95e6e500cb","id":"584633d4-7a8e-4c40-aa69-7a4851433c05","width":2406,"height":1258,"filename":"Screenshot 2024-03-31 at 6.41.47 PM.png","type":"image/png","caption":"","border":false},{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FUKLkhMFOJFbDx4g19Xqd%2F6b3d14b9-e10d-4f6a-8ae4-0029b7b71aae.png?alt=media&token=f235639b-5747-447f-a9a0-2ebe4d3be52d","id":"6b3d14b9-e10d-4f6a-8ae4-0029b7b71aae","width":2398,"height":1236,"filename":"Screenshot 2024-03-31 at 6.41.27 PM.png","type":"image/png","caption":"","border":false},{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FUKLkhMFOJFbDx4g19Xqd%2Fb00144e4-cece-42fd-ac20-cfeaf2c38a78.png?alt=media&token=676bb733-2a1a-4692-9e22-a91e9825157f","id":"b00144e4-cece-42fd-ac20-cfeaf2c38a78","width":2412,"height":1266,"filename":"Screenshot 2024-03-31 at 6.41.09 PM.png","type":"image/png","caption":"","border":false}]}
```

NYU Campus Redesign

In the class Design for Climate Change, Professor Peder Anker gave us a simple prompt:

*Turn Trash into Treasure.*

My goal for this assignment was to create something that was practical and usable, and I decided to build a chair. A comfortable chair that could hold my body weight would be a quintessentially practical and useful object.

Materials included wood beams from a bed frame, pillow case, piping, plastic bottles, rags, non-slip mat.

To turn the plastic bottles into a sort of plastic string,  I built a special tool out of a cutting board, a thumb tack and a pencil sharpener.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FWolybRKpeMnWjva1VxK9%2F2169947e-c216-45c9-a5e7-f2dc03abcaa8.jpg?alt=media&token=d9044f0d-863a-4702-972b-387d521f41d0","id":"2169947e-c216-45c9-a5e7-f2dc03abcaa8","width":3024,"height":4032,"filename":"IMG_6486.jpg","type":"image/jpeg","caption":"","border":false}]}
```

The plastic string provided an adhesive to construct the chair. Wrapping the string around 2 components creates a weak bond, but when heat is applied the plastic melts slightly to conform to the curves and corners of the joint. When cooled the bond becomes extremely strong.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FWolybRKpeMnWjva1VxK9%2F093e5611-2f28-4823-bab1-a869e7500aca.jpg?alt=media&token=9061741b-6612-469c-aa38-f947388f4a70","id":"093e5611-2f28-4823-bab1-a869e7500aca","width":3024,"height":4032,"filename":"IMG_6492.jpg","type":"image/jpeg","caption":"","border":false}]}
```

I felt the actual design of the chair should represent the materials it is composed of. This means the chair should be simple, lightweight and easily transported. Additionally, The aesthetics should reflect its construct as well; as you can see, there is no attempt to disguise the chair as anything it is not.

I was inspired by Victor Papanek’s “Lean-on chair” from his book **Nomadic Furniture**. His chair was designed to be extremely simple, use the least materials as possible, and be extremely portable.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FWolybRKpeMnWjva1VxK9%2F9b034270-0178-410e-b67e-8b4a33a7ad4e.jpg?alt=media&token=80fbbebe-975b-4e48-a8bc-87ea661fc4d2","id":"9b034270-0178-410e-b67e-8b4a33a7ad4e","width":3024,"height":4032,"filename":"Chair 1.jpg","type":"image/jpeg","caption":"","border":false},{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FWolybRKpeMnWjva1VxK9%2F8c73381f-13c1-495d-9997-6178af171709.jpg?alt=media&token=bc03355b-60b1-4481-87ac-0ca99c56ce2a","id":"8c73381f-13c1-495d-9997-6178af171709","width":3024,"height":4032,"filename":"Chair 2.jpeg","type":"image/jpeg","caption":"","border":false}]}
```

My design creates sort of a hammock with the pillow case, whose depth can be adjusted by changing the height of the bottom plank. Additionally, the user’s sitting angle can be adjusted by changing the angle as which the chair leans against the wall. The rags are employed as a cushion to prevent the corners from scratching the wall, and the non-slip mat keeps the bottom from sliding on the floor.

Trash Chair

For my Human Computer Interaction Final project, I wanted to improve a tool that I have used frequently and think is important, so I chose **Khan Academy.** To see the full deck **[click here](https://pitch.com/v/khan-ux-design-plan-ph92i5).**

### **Challenge:**

Khan Academy is a great resource, in part because of the depth and breadth of its available content. Khan Academy has content ranging from pre-k to college levels, and even ‘life skills’ courses, in over 30 different languages. This amount of content is amazing for their users but it can also be overwhelming. I set out to create a tool to help users better navigate through this content.

**Steps Taken:**

1. Understanding Khan Academy

2. Understanding the User

3. Design Motivation

4. Design Prototypes

**Understanding Khan Academy**

Khan Academy's mission statement is to “Provide a free world class education for anyone, anywhere”. Khan Academy’s curricula are based on a ‘Mastery Learning’ system. Mastery Learning simply means allowing a student to continue to work on a concept until they can master that concept or skill. This requires personalized curricula, based on students gaps in understanding. They target "unfinished learners", or students who haven’t mastered critical core skills. Khan Academy believes that learning is a aggregate, one level cannot be reached without total comprehension of the previous. Khan Academy also emphasizes students who wouldn’t normally have the resources to get a complete and world-class education.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F56c23b0a-8f5c-473c-8adf-eac057e02604.png?alt=media&token=701f1a69-a4a8-436b-a8e1-f9c9f7b4a8c6","id":"56c23b0a-8f5c-473c-8adf-eac057e02604","width":1792,"height":996,"filename":"Screenshot 2024-04-01 at 10.55.41 AM.png","type":"image/png","caption":"","border":false}]}
```

**Understanding the User**

I can break down the users of Khan Academy(the learners that is, what I am focusing on) into 2 groups:

*Using in conjunction with their schools, organized by the teacher or district*

* The instructor is facilitating the use of its materials, and guiding students along the process.

*Using independently of a school, working alone*

These groups can be further divided into those who use it consistently and those who don’t.

* Consistent Users, or Monthly Very Active Learners as Khan Academy calls them, are users who are following and completing courses in whole. Whether facilitated by an outside instructor or not, their aim is to achieve mastery in a subject.

* Pick-and-choose users, are users that are looking for support in specific areas;

* selecting 1 or just a small set of videos on a specific topic.

* they were left confused by a class in school, or want to review basic concepts for a more advanced subject.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F13be5886-6a4b-409c-919c-2f44346a747a.png?alt=media&token=bb65328e-ae82-44d4-ab8e-2410fa14b6ce","id":"13be5886-6a4b-409c-919c-2f44346a747a","width":1806,"height":1004,"filename":"Screenshot 2024-04-01 at 10.47.58 AM.png","type":"image/png","caption":"","border":false}]}
```

I developed a sample user persona to better understand one of Khan’s user groups.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F4a841660-4900-4847-9220-65860dc7384b.png?alt=media&token=db383008-2c73-4ff0-9cbf-84e684d4e08e","id":"4a841660-4900-4847-9220-65860dc7384b","width":1994,"height":1120,"filename":"Screenshot 2024-04-01 at 10.26.56 AM.png","type":"image/png","caption":"","border":false}]}
```

**Motivating the Design**

Knowing about the goals of Khan Academy as well as the user will motivate my design decisions.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2Fda413cc7-851b-4ad9-916e-506fd4eddd1d.png?alt=media&token=9d0c4093-edbd-420e-9049-50bdb29df272","id":"da413cc7-851b-4ad9-916e-506fd4eddd1d","width":1790,"height":992,"filename":"Screenshot 2024-04-01 at 10.58.53 AM.png","type":"image/png","caption":"","border":false}]}
```

I also found where the Khan Academy UX comes up short in providing for its users.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F62c55c2d-e358-4098-8c11-8aff77e040f4.png?alt=media&token=f2b6aad2-f717-46af-81d3-50e1a0ad3f22","id":"62c55c2d-e358-4098-8c11-8aff77e040f4","width":1800,"height":1000,"filename":"Screenshot 2024-04-01 at 11.01.57 AM.png","type":"image/png","caption":"","border":false}]}
```

Let’s say the user selects **add new course**. They are met with this pop-up window, where they can choose from Khan Academy’s long list of courses. This might be fine if you knew the exact name of the course you wanted to take, but if users aren’t being told what course to select, it may not line up with their other education source. And what about unfinished learners? If you’re in 9th grade math, but have gaps in your previous math knowledge how will Khan Academy help you achieve Mastery Learning? What if you want to start from the middle of a course? Or a specific topic? That’s the drawback of having so much content, having to filter through it. If you were to click on a subject you would have to attempt to search through litany of courses, and the numerous units within each course.

Or, you could try the **Search function.** Look at the results for the simple search term ‘derivative’. First of all, the results are not correct, a student wanting to learn about derivatives isn’t going to be helped with content about second derivatives. 2nd, all the types of content Khan Academy produces are shown together: lessons, exercises, videos, articles. This is very ambiguous for a user. Where do you start? Do you click on lessons, videos, exercises, or articles? It's not clear at all. And remember that unfinished learners **don’t know what they don’t know.** That will only add to the ambiguity. For instance, they are struggling with derivatives, but they have never heard of a riemann sum, or the fundamental theorem of calculus.

**Designing Prototypes**

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F275f16e0-d9e3-4916-a6fb-afd1bbb5cb85.png?alt=media&token=8b6fda98-1c29-40cb-9d2b-5f2e829ead81","id":"275f16e0-d9e3-4916-a6fb-afd1bbb5cb85","width":1796,"height":976,"filename":"Screenshot 2024-04-01 at 11.09.54 AM.png","type":"image/png","caption":"","border":false}]}
```

With these motivations in mind, I've made a number of changes to the dashboard:

First I've made the search bar bigger and more prominent, to more clearly signify the search function. I've also made the **Resume Learning** videos more discoverable.

The new course panels have a visual language that heavily emphasize progress, making the system status clear. Right away the user understands how much they've completed of each unit, without even having to read the percentages, the progress bars indicate how far the user has gotten. The layout of the course panel also reflects the ideas of Mastery Learning. If users move through the units without having fully mastered the content, the bars will appear unbalanced, indicating unfinished learning, and gaps in knowledge.

In addition to redesigning the dashboard, I designed a new user journey for starting a new learning topic, using AI.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F98835c18-30eb-4151-baf5-4c9ae6404a5e.png?alt=media&token=bc737095-185f-454e-b80c-6587f7ea012f","id":"98835c18-30eb-4151-baf5-4c9ae6404a5e","width":1794,"height":1008,"filename":"Screenshot 2024-04-01 at 11.14.36 AM.png","type":"image/png","caption":"","border":false}]}
```

Right now Khan Academy is leveraging AI technology to create a one-on-one tutor, called *Khanmigo* for users: answering their questions, giving suggestions, prompting deeper thinking, basically just facilitating deeper learning. But I think their AI tool has even more potential, specifically as a way to help users find the content they need. This will help users bridge that gulf of execution by directing them to a starting place, rather than forcing them to find it on their own.

I created a paper prototype to test the design:

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2F3995fb97-f3b6-4afb-a32e-be773d324193.mov?alt=media&token=70e3cb56-56fa-40b6-ae5d-8554f8111f7a","id":"3995fb97-f3b6-4afb-a32e-be773d324193","width":360,"height":640,"filename":"paper prototype.mov","type":"video/quicktime","caption":"","border":false}]}
```

And a wireframe:

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FXbjF8VsA9XJkAJC36A6Q%2Fa3461121-52da-4051-8f8e-9bcf5c228ac6.png?alt=media&token=1721cac1-2630-4eb8-9249-ab81bfb605ae","id":"a3461121-52da-4051-8f8e-9bcf5c228ac6","width":2356,"height":1276,"filename":"Screenshot 2024-04-05 at 6.29.48 PM.png","type":"image/png","caption":"","border":false}]}
```

These proposed UX changes make it easier for users to find the information they need and in doing so would further Khan Academy's mission of creating Mastery Learning in their users.

Khan Academy UX Design Plan

To see the entire deck [click here](https://docs.google.com/presentation/d/e/2PACX-1vSfdz2NzWq8FZAfL8pFs5lq3S08mW-9zQVlKGpg4ZZoBTU0JVrdYSUgPRdLlgkj8yYdHPpjXXJi30Ip/pub?start=false&loop=false&delayms=3000).

As society continues to pollute and trash the earth, the frequency of extreme weather events is continually rising. Climate change presents the most pressing issue of our time. As extreme weather events become more intense and more frequent, being able to detect extreme weather is and will become crucial.

Using ML to predict weather is not a novel idea, but most publications relating to weather forecasting don’t focus on **extreme weather**. Because of this, extreme weather events are often missed in their forecasts (explained in Watson’s paper *[Machine learning applications for weather and climate need greater focus on extremes)](https://iopscience.iop.org/article/10.1088/1748-9326/ac9d4e).*

My goal with this project was to train various ML models to predict extreme weather events. Most weather prediction models can be considered regression tasks, because they aim to predict a continuous variable (something like temperature, or rainfall), but I formulated my problem into a classification task. The model will predict either a 1, meaning a day will have extreme weather, or 0, meaning a day will not have extreme weather.

The dataset I used was a NOAA daily climate summary from NYC, from Jan 1, 1980 to Jan 1, 2023.  As you will see I do a lot of preprocessing and feature engineering to this data.

First, I need to build a target vector (or the ground truth, i.e. what is 'extreme weather'). I decided for the first version I would choose temperature as the measure of extreme weather (i.e. extreme heat, or extreme cold).  I used Season-trend-loess decomposition to find days of extreme heat. STL decomposition is a method of breaking down time series data into 3 components: Trend, Seasonal, and Residual. It is very effective for data that is cyclical, like climate.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2F1f1f9303-27af-410c-92df-6c6b4c5adcf3.png?alt=media&token=16a69462-930a-4df0-8d64-03df28a4a121","id":"1f1f9303-27af-410c-92df-6c6b4c5adcf3","width":1235,"height":988,"filename":"stl decomp.png","type":"image/png","caption":"","border":false}]}
```

The Residuals component of STL represents the data that cannot be accounted for by either the seasonal component or the overall trend component. We can find anomalies in the data by looking at residuals that lie outside 3 STD of the mean, I will call those days with extreme weather.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2F75d0025d-4e08-47e1-9037-c4683c50685d.png?alt=media&token=2f4bae2f-d8e6-4559-944c-9afc4ece59e7","id":"75d0025d-4e08-47e1-9037-c4683c50685d","width":2024,"height":874,"filename":"outlier.png","type":"image/png","caption":"","border":false}]}
```

I have a feature vector (a bunch of climate data, rainfall, humidity, wind speed, etc.) and a target vector (which days have extreme temperatures), but now I need to make the model *predictive*. To be useful it has to be able to tell days in advance whether a day will have extreme weather. To do this I can use **Feature Lag.**  Feature lag means shifting the feature vector in time, so that a given target date(say April 10th) is looking at data from previous days ( say April  7th, 6th, and 5th). For this model I used lags of 4-10, meaning the model only looks at data from 4-10 days prior.

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2F939fa7e4-9a47-48ea-903d-d4b7accb0b97.png?alt=media&token=55aa71ba-a052-4a04-94d1-f5bf54d8a332","id":"939fa7e4-9a47-48ea-903d-d4b7accb0b97","width":1202,"height":788,"filename":"Screenshot 2024-04-01 at 12.55.43 PM.png","type":"image/png","caption":"","border":false}]}
```

**Now I can train some models!**

The models I used to test are:

1. Decision Trees

2. Random Forest

3. Adaboost

4. Support Vector Machine

I choose these models because they are classifiers that function well under high dimensional space, can handle nonlinear relationships between classes, and they are interpretable, meaning I can find out which data points are most influential.

After training, I got some promising results,

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2Ff490a05c-cd92-4316-a120-b659c112a262.png?alt=media&token=aca58669-024e-416a-b688-2995de07f895","id":"f490a05c-cd92-4316-a120-b659c112a262","width":1136,"height":954,"filename":"Screenshot 2024-04-01 at 1.01.56 PM.png","type":"image/png","caption":"","border":false}]}
```

These are great results for a first test, but we can do better.

**I want to add some features to the dataset, specifically:**

**Rolling window:** selecting a period (i.e. 7 days) and calculating the mean, like a moving average;

**Expanding window:** a cumulative window, as more samples are read they are added to the window and the mean is calculated; we can do it per month and per day. So a month average is calculated for each preceding month (the same month though, so all the past junes for example). Same for the daily expanding average;

Both of these features serve to incorporate more past data into any given day's prediction.

Lets see how it works!

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2F02b7eab6-9e48-4856-a471-48c7519b925c.png?alt=media&token=cf06b6b9-45bc-4eec-8431-59e97e8bf234","id":"02b7eab6-9e48-4856-a471-48c7519b925c","width":1142,"height":918,"filename":"Screenshot 2024-04-01 at 1.05.52 PM.png","type":"image/png","caption":"","border":false}]}
```

Wow, the new dataset actually performed worse! I think that is because the first dataset has more features (161 to 66, because I got rid of all the normal lags). It's also interesting that the same models are more successful for both sets.

But this is just the beginning!

**Some things I am going to explore in the future:**

1. **Creating a Physics-informed loss function**

   Loss functions are how machine learning systems learn. They are a way to measure whether the algorithm is doing a good job by determining how far away the current output is from the expected output.  We know that the climate follows certain physical laws, gases have to behave a certain way, heat rises, rain falls blah blah blah. By incorporating these laws into the loss function, we reduce the output space(the number of possible learned outputs) by  ensuring that the solution learned by the network is consistent with the known physics. More [here](https://benmoseley.blog/my-research/so-what-is-a-physics-informed-neural-network/)

2. **How can I incorporate spatial relationships as well as temporal ones?**

   It's safe to say that weather in one area is affected by weather in the surrounding areas, so by inputting more data about the weather in different locations, my model will be more accurate. By using neural networks I can have inputs with high dimensionality, like images. Instead of just having numeric data, these images can convey data with color and brightness, like this:

```cv
gallery
{"layout":"hscroll","images":[{"src":"https://firebasestorage.googleapis.com/v0/b/maitake-project.appspot.com/o/pages%2FqaTAv7FyxqQ02SVzTrQo9PARQR03%2FpuvmaFiOajEXwuf0IpSy%2Ffae9889f-cfe0-4e58-be29-b506bf08c428.png?alt=media&token=77285d63-bc41-497d-958b-f4869ad2112a","id":"fae9889f-cfe0-4e58-be29-b506bf08c428","width":844,"height":628,"filename":"Screenshot 2024-04-01 at 1.12.58 PM.png","type":"image/png","caption":"","border":false}]}
```

Forecasting Extreme Weather with ML

Masters Student

BA Physics, Minors  in Math and Computer Science, and Science and Technology Studies

Forecasting Extreme Weather

Khan Academy UX Design

Trash chair

* Assisted History and Philosophy of Science Professor Matthew Stanley with research for a book

* Found primary and secondary sources, summarized and  cataloged them in a detailed reference library

* Communicate with the Library of Congress and various institutions’ archivists

Research Assistant

* Researched what drives Generation Z consumer and social media behavior.

* Developed metrics to describe and analyze that consumer behavior.

* Create methods to help companies connect with GenZ

* Clients include: Denny’s, Nonfiction Research, Atom, Yunit

Research and Strategy Intern

* Research and development of a UI/UX to visualize and parse social media metrics

* Analyzed social media data, and applied statistical models and physics concepts to develop a metric for measuring the momentum of cultural terms

Research and Development

* Communicate with Strategy and Design departments to complete client briefs

* Build web apps and components using Javascript and ReactJS

* Researched and Designed B2B interface for a Fortune 500 company as a part of a capstone project

Ian A

2024