Category

Research

Using GenAI to design floor plans and buildings

By | Data, Data sets, Educational, Feature, General Interest, HPC, News, Research, Systems and Services

There is a lot to consider when designing places where humans live and work. How will the space be used? Who’s using the space? What are budget considerations? It is painstaking and time consuming to develop all of those details into something usable. 

What if Generative AI (GenAI) could help? We already know that it can be used to create text, music, and images. Did you know that it can also create building designs and floor plans? 

Dr. Matias del Campo, associate professor of architecture in the Taubman College for Architecture and Urban Planning, has been working to make architectural generative models more robust. He aims to expand on the patterns, structures, and features from the available input data to create architectural works. Himself a registered architect, designer, and educator, del Campo conducts research on advanced design methods in architecture using artificial intelligence techniques.

He leverages something called neural networks for two projects: 

  • Common House: A project that focuses on floor plan analysis and generation.
  • Model Mine: A large-scale, 3D model housing database for architecture design using Graph Convolutional Neural Networks and 3D Generative Adversarial Networks.

This is an example from the annotated data created from the Common House research project. The main obstacle that has emerged in creating more real-life plans is the lack of databases that are tailored for these architecture applications. The Common House project aims at creating a large-scale dataset for plans with semantic information. Precisely, our data creation pipeline consists of annotating different components of a floor plan, for e.g., Dining Room, Kitchen, Bed Room, etc.

 

Four quadrants showing 9 models each of chairs, laptops, benches, and airplanes

A large scale 3D model housing database for Architecture design using Graph Convolutional Neural Networks and 3D Generative Adversarial Networks.

What exactly are neural networks? The name itself takes inspiration from the human brain and the way that biological neurons signal to one another. In the GenAI world, neural networks are a subset of machine learning and are at the heart of deep learning algorithms. This image of AI hierarchy may be helpful in understanding how they are connected.

Dr. del Campo’s research uses GenAI for every step of the design process including 2D models for things like floors and exteriors, and 3D models for shapes of the rooms, buildings, and volume of the room. The analysis informs design decisions. 

DEI considerations

del Campo notes that there are some DEI implications for the tools he’s developing. “One of the observations that brought us to develop the ‘Common House’ (Plangenerator) project is that the existing apartment and house plan datasets are heavily biased towards European and U.S. housing. They do not contain plans from other regions of the world; thus, most cultures are underrepresented.” 

To counterbalance that, del Campo and his team made a global data collection effort, collecting plans and having them labeled by local architects and architecture students. “This not only ensured a more diverse dataset but also increased the quality of the semantic information in the dataset.”

How technology supports del Campo’s work

A number of services from Information Technology & Services are used in these projects, including: Google at U-M collaboration tools, GenAI, Amazon Web Services at U-M (AWS), and GitHub at U-M

Also from ITS, the Advanced Research Computing (ARC) team provides support to del Campo’s work. 

“We requested allocations from the U-M Research Computing Package for high-performance computing (HPC) services in order to train two models. One focuses on the ‘Common House’ plan generator, and the other focuses on the ‘Model Mine’ dataset to create 3D models based,” said del Campo. 

Additionally, they used HPC allocations from the UMRPC in the creation of a large-scale artwork called MOSAIK which consists of over 20,000 AI-generated images, organized in a color gradient. 

A large scale 3D model housing database for Architecture design using Graph Convolutional Neural Networks and 3D Generative Adversarial Networks.

“We used HPC to run the algorithm that organized the images. Due to the necessary high resolution of the image, this was only possible using HPC.”

“Dr. del Campo’s work is really novel, and it is different from the type of research that is usually processed on Great Lakes. I am impressed by the creative ways Dr. del Campo is applying ITS resources in a way that we did not think was possible,” said Brock Palen, director of the ITS Advanced Research Computing. 

Related: Learn about The Architecture + Artificial Intelligence Laboratory (AR2IL)

Using natural language processing to improve everyday life

By | Data, Great Lakes, HPC, News, Research, Uncategorized

Joyce Y. Chai, professor of electrical engineering and computer science, College of Engineering, and colleagues have been seeking answers to complex questions using natural language processing and machine learning that may improve everyday life.

Some of the algorithms that they develop in their work are meant for tasks that machines may have little to no prior knowledge of. For example, to guide human users to gain a particular skill (e.g., building a special apparatus or even, “Tell me how to bake a cake”). A set of instructions based on the observation of what the user is doing, e.g., to correct mistakes or provide the next step, would be generated by Generative AI, or GenAI. The better the data and engineering behind the AI, the more useful the instructions will be. 

“To enable machines to quickly learn and adapt to a new task, developers may give a few examples of recipe steps with both language instructions and video demonstrations. Machines can then (hopefully) guide users through the task by recognizing the right steps and generating relevant instructions using GenAI,” said Chai.

What are AI, machine learning, deep learning, and natural language processing?

It might help to take a step back to understand AI, machine learning (ML), and deep learning at a high level. Both ML and deep learning are subsets of AI, as seen in the figure. Some natural language processing (NLP) tasks fall within the realm of deep learning. They all work together and build off of each other.

Artificial Intelligence, or AI, is a branch of computer science that attempts to simulate human intelligence with computers. It involves creating systems to perform tasks that usually need human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

“NLP is a sub-area in AI, where DL/ML approaches are predominantly applied,” stated Chai.

Christopher Brown, a research data scientist, with ITS Advanced Research Computing (ARC) and a member of the ARC consulting team, explains that ML is a subfield of AI. Within ML, algorithms are used to generalize situations beyond those seen in training data and then complete tasks without further guidance from people. A good example is U-M GPT. The large language models (LLMs) accessible via U-M GPT are trained with millions of diverse examples. “The goal is to train the models to reliably predict, translate, or generate something.” 

“Data is any information that can be formatted and fed into an algorithm that can be used for some task, including journal articles, chats, numbers, videos, audio, and texts,” said Brown, Algorithms can be trained to perform tasks using these real-world data. 

Natural Language Processing is a branch of artificial intelligence that helps computers understand and generate human language in a way that is both meaningful and useful to humans. NLP teaches computers to understand languages and then respond so that humans can understand, and even accounting for when rich context language is used. 

“NLP is highly interdisciplinary, and involves multiple fields, such as computer science, linguistics, philosophy, cognitive science, statistics, mathematics, etc.,” said Chai.

Examples of NLP are everywhere: when you ask Siri for directions, or when Google efficiently completes your half-typed query, or even when you get suggested replies in your email. 

Ultimately NLP, along with AI, can be used to make interactions between humans and machines as natural and as easy as possible. 

A lot of data is needed to train the models

Dr. Chai and her team use large language models, a lot of data, and computing resources. These models take longer to train and are harder to interpret. Brown says, “The state of the art, groundbreaking work tends to be in this area.” 

Dr. Chai uses deep learning algorithms that make predictions about what the next part of the task or conversation is. “For example, they use deep learning and the transformer architecture to enable embodied agents to learn how new words are connected to the physical environment, to follow human language instructions, and to collaborate with humans to come up with a shared plan,” Brown explains.

The technology that supports this work

To accomplish her work, Dr. Chai uses the Great Lakes High-Performance Computing Cluster and Turbo Research Storage, both of which are managed by U-M’s Advanced Research Computing Group (ARC) in Information and Technology Services. She has 16 GPUs on Great Lakes at the ready, with the option to use more at any given time. 

A GPU, or Graphics Processing Unit, is a piece of computer equipment that is good at displaying pictures, animations, and videos on your screen. The GPU is especially adept at quickly creating and manipulating images. Traditionally, GPUs were used for video games and professional design software where detailed graphics were necessary. But more recently, researchers including Dr. Chai discovered that GPUs are also good at handling many simple tasks at the same time. This includes tasks like scientific simulations and AI training where a lot of calculations need to be done in parallel (which is perfect for training large language models).

“GPUs are popular for deep learning, and we will continue to get more and better GPUs in the future. There is a demand, and we will continue supporting this technology so that deep learning can continue to grow,” said Brock Palen, ITS Advanced Research Computing director. 

Chai and her team also leveraged 29 terabytes of the Turbo Research Storage service at ARC. NLP benefits from the high-capacity, reliable, secure, and fast storage solution. Turbo enables investigators across the university to store and access data needed for their research via Great Lakes. 

Great Lakes HPC in the classroom 

ARC offers classroom use of high-performance computing cluster resources on the Great Lakes High-Performance Computing Cluster

Dr. Chai regularly leverages this resource. “Over 300 students have benefited from this experience. We have homework that requires the use of the Great Lakes, e.g., having students learn how to conduct experiments in a managed job-scheduling system like SLURM. This will benefit them in the future if they engage in any compute-intensive R&D (research and development).

“For my NLP class, I request Great Lakes access for my students so they have the ability to develop some meaningful final projects. We also use the Great Lakes HPC resources to study the reproducibility for NLP beginners,” said Chai. A gallery is available for many of the student projects.

The UMRCP defrays costs

The U-M Research Computing Package is a set of cost-sharing allocations offered by ITS ARC, and they are available upon request. Other units offer additional cost-sharing to researchers. Chai said, “We typically use the nodes owned by my group for research projects that require intensive, large-scale GPU model training. We use the UMRCP for less intensive tasks, thereby extending the budgetary impact of the allocations.” 

New Hardware-Accelerated Visualization Partition 

By | General Interest, Great Lakes, HPC, News, Research

We are excited to introduce a brand-new feature of Great Lakes to researchers who need hardware acceleration for their visualization requirements. ARC has created a specialized “viz” partition, consisting of four nodes equipped with NVIDIA P40 GPUs. These nodes are accessible through Open OnDemand’s Remote Desktop functionality.

Key details about this new feature:

  • Jobs utilizing the viz partition have a maximum walltime of 2 hours.
  • The charge rate for the viz partition is currently aligned with our standard partition. 

To make use of the viz partition, follow these steps:

  • Create a Remote Desktop job on Great Lakes via Open OnDemand.
  • Request the “viz” partition, specifying 1 node and 1 GPU (Please note that we have only one GPU available per node).
  • Prefix any application you intend to run with accelerated graphics with the command “vglrun.” (example: “vglrun glxgears”)

For questions or support requests, please contact our team at arc-support@umich.edu.

What DNA can tell us about dog evolution

By | General Interest, HPC, News, Research, Systems and Services

An excerpt from the Michigan Medicine Health Lab Podcast:

An international consortium of scientists, led by Jeff Kidd, Ph.D., of the University of Michigan, Jennifer R. S. Meadows of Uppsala University in Sweden, and Elaine A. Ostrander, Ph.D. of the NIH National Human Genome Research Institute, is using an unprecedentedly large database of canine DNA to take an unbiased look at how our furry friends evolved into the various breeds we know and love.

A paper, published in the journal Genome Biology, outlines what the Dog10K project discovered after sequencing the genomes of close to 2,000 samples from 321 different breed dogs, wild dogs, coyotes, and wolves, and comparing them to one reference sample—that of a German Shepherd named Mischka.

Analyzing more than 48 million pieces of genetic information, they discovered that each breed dog had around 3 million single nucleotide polymorphism differences.

These SNPs or “snips” are what account for most of the genetic variation among people and dogs alike.

They also found 26,000 deleted sequences that were present in the German Shepherd but not in the comparison breed and 14,000 that were in the compared breed but missing from Mischka’s DNA.

“We did an analysis to see how similar the dogs were to each other, and it ended up that we could divide them into around 25 major groups that pretty much match up with what people would have expected based on breed origin, the dogs’ type, size and coloration,” said Kidd.

Most of the varying genes, he added, had to do with morphology, confirming that the breed differences were driven by how the dogs look.

Relative to dogs, wolves had around 14% more variation. And wild village dogs—dogs that live amongst people in villages or cities but aren’t kept as pets—exhibited more genetic variation than breed dogs.

The data set, which was processed using the Great Lakes high-performing computing cluster at U of M, also revealed an unusual amount of retrogenes, a new gene that forms when RNA gets turned back into DNA and inserted back into the genome in a different spot.

Listen to the podcast or read the transcript on the Health Lab webpage.

Technology supports researchers’ quest to understand parental discipline behaviors

By | Feature, HPC, News, Research, Systems and Services, Uncategorized

Image by Rajesh Balouria from Pixabay

How do different types of parental discipline behaviors affect children’s development in low- and middle-income countries (LMICs)? A group of researchers set out to understand that question. They used a large data set from UNICEF of several hundred thousand families. The data came from the fourth (2009–2013) and fifth (2012–2017) rounds of the UNICEF Multiple Indicator Cluster Surveys. 

“The majority of parenting research is conducted in higher income and Westernized settings. We need more research that shows what types of parenting behaviors are most effective at promoting children’s development in lower resourced settings outside of the United States. I wanted to conduct an analysis that provided helpful direction for families and policymakers in LMICs regarding what parents can do to raise healthy, happy children,” said Kaitlin Paxton Ward, People Analytics Researcher at Google and Research Affiliate at the University of Michigan.

Dr. Paxton Ward is the lead author on the recently-released paper, “Associations between 11 parental discipline behaviors and child outcomes across 60 countries.” Other authors are also cited in the article: Andrew Grogan-Kaylor, Julie Ma, Garrett T. Pace, and Shawna Lee.

Together, they tested associations between 11 parental discipline behaviors and outcomes (aggression, distraction, and prosocial peer relations) of children under five years in 60 LMICs:

  • Verbal reasoning (i.e., explaining why the misbehavior was wrong)
  • Shouting
  • Name calling
  • Shaking
  • Spanking
  • Hitting/slapping the body
  • Hitting with an object 
  • Beating as hard as one could
  • Removing privileges 
  • Explaining
  • Giving the child something else to do

Results

Verbal reasoning and shouting were the most common parental discipline behaviors towards young children. Psychological and physical aggression were associated with higher child aggression and distraction. Verbal reasoning was associated with lower odds of aggression, and higher odds of prosocial peer relations. Taking away privileges was associated with higher odds of distraction, and lower odds of prosocial peer relations. Giving the child something else to do was associated with higher odds of distraction. The results indicated that there was some country-level variation in the associations between parenting behaviors and child socioemotional outcomes, but also that no form of psychological or physical aggression benefitted children in any country.

Conclusion 

Parental use of psychological and physical aggression were disadvantageous for children’s socioemotional development across countries. Only verbal reasoning was associated with positive child socioemotional development. The authors suggest that greater emphasis should be dedicated to reducing parental use of psychological and physical aggression across cultural contexts, and increasing parental use of verbal reasoning.

The technology used to analyze the data

The researchers relied on a complicated Bayesian multilevel model. This type of analysis incorporated knowledge from previous studies to inform the current analysis, and also provided a way for the researchers to look in more detail at variation across countries. To accomplish this task, the team turned to ITS Advanced Research Computing (ARC) and the Great Lakes High-Performance Computing Cluster. Great Lakes is the largest and fastest HPC service on U-M’s campus. 

“I know for me as a parent of young children, you want the best outcome. I have known people to grow up with different forms of discipline and what the negative or positive influence of those are,” said Brock Palen, ARC director. 

The researchers also created a visual interpretation of their paper for public outreach using a web app called ArcGIS StoryMaps. This software helps researchers tell the story of their work. With no coding required, StoryMaps combine images, text, audio, video, and interactive maps in a captivating web experience. StoryMaps can be shared with groups of users, with an organization, or with the world. 

All students, faculty, and staff have access to ArcGIS StoryMaps. Since 2014, U-M folks have authored over 7,500 StoryMaps, and the number produced annually continues to increase year-over-year. Explore examples of how people around the world are using this technology in the StoryMaps Gallery.

“This intuitive software empowers the U-M community to author engaging, multimedia, place-based narratives, without involving IT staff,” said Peter Knoop, research consultant with LSA Technology Services. 

Correspondence to Dr. Kaitlin Paxton Ward, kpward@umich.edu.

Related article

HPC Emergency 2023 Maintenance: September 15

By | Data, HPC, News, Research, Systems and Services

Due to a critical issue which requires an immediate update, we will be performing updates to Slurm and underlying libraries which allow parallel jobs to communicate. We will be updating the login nodes and the rest of the cluster on the fly and you should only experience minimal impact when interacting with the clusters. 

  • Jobs that are currently running will be allowed to finish. 
  • All new jobs will only be allowed to run on nodes which have been updated. 
  • The login and Open OnDemand nodes will also be updated, which will require a brief interruption in service.

Queued jobs and maintenance reminders

Jobs will remain queued, and will automatically begin after the maintenance is completed. Any parallel using MPI will fail; those jobs may need to be recompiled, as described below. Jobs not using MPI will not be affected by this update.

Jobs will be initially slow to start, as compute nodes are drained of running jobs so they can be updated. We apologize for this inconvenience, and want to assure you that we would not be performing this maintenance during a semester unless it was absolutely necessary.

Software updates

Only one version of OpenMPI (version 4.1.6) will be available; all other versions will be removed. Modules for the versions of OpenMPI that were removed will warn you that it is not available, as well as prompt you to load openmpi/4.1.6. 

When you use the following command, it will default to openmpi/4.1.6:
module load openmpi 

Any software packages you use (provided by ARC/LSA/COE/UMMS or yourself) will need to be updated to use openmpi/4.1.6. The software package updates will be completed by ARC. The code you compile yourself will need to be updated by you.

Note that at the moment openmpi/3.1.6 will be discontinued and warned to update your use to openmpi/4.1.6.

Status updates

 

System software changes

Great Lakes, Armis2 and Lighthouse

NEW version in BOLD OLD version
Slurm 23.02.5 compiles with:
  • PMIx
    • /opt/pmix/3.2.5
    • /opt/pmix/4.2.6
  • hwloc 2.2.0-3 (OS provided)
  • ucx-1.15.0-1.59056 (OFED provided)
  • slurm-libpmi
  • slurm-contribs
Slurm 23.02.3 compiles with:
  • PMIx
    • /opt/pmix/2.2.5
    • /opt/pmix/3.2.3
    • /opt/pmix/4.2.3
  • hwloc 2.2.0-3 (OS provided)
  • ucx-1.15.0-1.59056 (OFED provided)
  • slurm-libpmi
  • slurm-contribs
PMIx LD config /opt/pmix/3.2.5/lib PMIx LD config /opt/pmix/2.2.5/lib
PMIx versions available in /opt :
    • 3.2.5
    • 4.2.6
PMIx versions available in /opt :
  • 2.2.5
  • 3.2.3
  • 4.1.2
OpenMPI
  • 4.1.6
OpenMPI
  • 3.1.6
  • others

 

How can we help you?

For assistance or questions, contact ARC at arc-support@umich.edu.

Summer 2023 Network Maintenance: HPC and storage unavailable August 21-22 

By | Data, HPC, News, Research, Systems and Services

During the 2023 summer maintenance, a significant networking software bug was discovered and ARC was unable to complete the ARC HPC and Storage network updates at the MACC Data Center.

ITS has been working with the vendor on a remediation, and it will be implemented on August 21-22.  This will require scheduled maintenance for the HPC clusters Great Lakes, Armis2, and Lighthouse, as well as the ARC storage systems Turbo, Locker, and Data Den. The date was selected to minimize any impact during the fall semester. 

Maintenance dates:

HPC clusters and storage systems (/home and /scratch) and ARC storage systems (Turbo, Locker, and Data Den) will be unavailable August 21 starting at 7:00am.  Expected completion date is August 22nd.

Queued jobs and maintenance reminders

Jobs will remain queued, and will automatically begin after the maintenance is completed. The command “maxwalltime” will show the amount of time remaining until maintenance begins for each cluster, so you can size your jobs appropriately. The countdown to maintenance will also appear on the ARC homepage

Status updates

How can we help you?

For assistance or questions, contact ARC at arc-support@umich.edu.

U-M Research Computing Package automatic renewal begins July 1

By | News, Research, Uncategorized

** Looking for the LARCC Application?

____

The no-cost bundle of supercomputing resources known as the U-M Research Computing Package (UMRCP) automatically renews for most on July 1. 

Provided by Information and Technology Services, the UMRCP offers qualified researchers on all campuses (Ann Arbor, Dearborn, Flint, and Michigan Medicine) with allocations of high-performance computing, secure enclave, and research storage services. (Many units, including Michigan Medicine, provide additional resources to researchers. Be sure to check with your school or college.) 

If a faculty researcher has left the university (or is about to), and their research remains at the university, an alternative administrator must be assigned via the ARC Resource management Portal (RMP) so that the allocations can continue uninterrupted. ARC is available to help researchers make this transition. 

Don’t have the UMRCP? Here’s how to request resources 

Faculty, as well as staff and PhD students with their own funded research on all campuses (Ann Arbor, Dearborn, Flint, and Michigan Medicine), are welcome to request allocations. Full details are available on the Advanced Research Computing website

PhD researchers who do not have their own funded research can work with their advisor to be added to their allocations via the ARC Resource Management Portal (RMP).

“The UMRCP was launched in 2021 to meet the needs of a diversity of disciplines and to provide options for long-term data management, sharing, and protecting sensitive data,” said Brock Palen, director, ITS Advanced Research Computing. “The UMRCP alleviates a lot of the pressure that researchers feel in terms of managing the technology they need to achieve breakthroughs.”

More information

Globus maintenance happening at 9 a.m. on March 11

By | Armis2, Data, General Interest, Great Lakes, HPC, News, Research, Uncategorized

Due to planned maintenance by the vendor, Globus services will be unavailable for up to two hours beginning at 9 a.m. U.S. Eastern Time (10 a.m. Central Time) on Saturday, March 11, 2023.

Customers will not be able to authenticate or initiate any transfers during that time. Any transfers that have started before the outage will be stalled until the outage is over. Transfers will resume once maintenance is complete.

More details are available on the Globus blog.

For assistance or questions, please contact ARC at arc-support@umich.edu.

Dailey receives U-M Robotics’ first-ever alumni award 

By | General Interest, Happenings, News, Research, Uncategorized

Meghan Dailey will be presenting The Future of Machine Learning in Robotics on September 23 at 2 p.m., at FMCRB or Zoom

Meghan Dailey is the U-M Robotics department’s first Alumni Merit Award recipient!

Dailey is a member of the first-ever class in U-M Robotics. She earned a Masters of Science degree in 2015 with a focus in artificial intelligence. She is currently a machine learning specialist with Advanced Research Computing (ARC), a division of Information and Technology Service (ITS)

You’re invited 

In honor of the award, Dailey will be presenting “The Future of Machine Learning in Robotics” on Friday, September 23, 2 p.m., Ford Robotics Building (FMCRB) or on Zoom (meeting ID: 961 1618 4387, passcode: 643563). Machine learning is becoming widely prevalent in many different fields, including robotics. In a future where robots and humans assist each other in completing tasks, what is the role of machine learning, and how should it evolve to effectively serve both humans and robots? Dailey will discuss her past experiences in robotics and machine learning, and how she envisions machine learning contributing to the growth of the robotics field.

About Dailey

A member of the ARC Scientific Computing and Research Consulting Services team, Dailey helps researchers with machine learning and artificial intelligence programming. She has consulted with student and faculty teams to build neural networks for image analysis and classification. She also has extensive experience in natural language processing and has worked on many projects analyzing text sentiment and intent.

Image courtesy U-M Robotics