Showing posts with label science. Show all posts
Showing posts with label science. Show all posts

No posts for a wee while

Mike's Notes

I was on holiday for the last few weeks and am back now. There will be no blog posts, newsletters or meetings until Pipi Core is back up and running.

Update 27/05/2026

Lots of surprises. Making rapid progress. The peace and quiet are bliss.

Update 31/05/2026

The problem and solution are how things are named. Pipi auto-generates thousands of code names using multiple pattern languages, and all the naming conventions require many minor fixes for several unexpected reasons after migrating from a developer laptop to a production server environment. Everything else is absolutely fine.

Other naming problems are also being solved now, including:

  • The rapid development of Boxlang by Ortus has brought forward another challenge. Pipi 10 will be migrated to run on top of Boxlang in 2027 to support multiple languages, including C++, CFML, COBOL, Go, Java, JavaScript, PHP, Python, Rust, etc.
  • Future integration with cloud-based LLMs.
  • Future integrations with Office365, Google Workspace, Zoho, LibreOffice, etc.

The common solution is to create standardised naming systems that are simple, stable, robust, schema-based, versioned, self-documenting, and extensible to meet unanticipated future needs.

This is done by replacing code-based naming rules with database-driven ones that can be easily edited in the future via an admin UI.

90% of these names are internal, hidden in the closed core, and how they work and what they are will not be discussed here. The rest will be publicly and fully documented as part of the open-source workspaces for developers to work with.

Update 02/06/2026

I'm changing the disclosure boundary between the Pipi closed-core and open-source workspaces. Previously, "disclose everything unless there is a security reason not to". This is now changed to "disclose on the basis of need to know".

Closed-core accounts for 90% and open-source workspaces for 10% of lines of code, databases, etc.

This will reduce the documentation burden, given Pipi's vast scale. So, the open-source workspaces will be fully shared and documented on GitHub, etc, without restriction. This includes;

  • Standards schema
  • Ontologies
  • Parameters
  • Laws of physics
  • HTML + CSS
  • Algorithms
  • Module DDD models
  • Workflow diagrams
  • Documentation
  • API schema
  • UI code
  • etc

This also means some existing technical documentation about the closed-core will become hidden and only available internally.

Update 07/06/2026

Pipi Core is the IDE used to edit Pipi Core (AKA: which came first, the chicken or the egg?). Temporary UIs have been created and are being used across multiple engines to edit the names in use. This is much faster than directly editing data, which had to be done initially. The next step will be turning auto-generation back on. Once that's done, temporary UIs will be used to build permanent UIs. More automation will then be enabled via the UIs, and so on, as Pipi Core builds itself with a human in the loop.

Update 08/06/2026

The list of code cases available to use now for auto-generated naming, I/O translation, etc with examples, includes;

  • camelCase: userProfilePicture
  • kebab-case: user-profile-picture
  • PascalCase: UserProfilePicture
  • snake_case: user_profile_picture
  • SCREAMING_SNAKE_CASE: USER_PROFILE_PICTURE
  • Train-Case: User-Profile-Picture
  • flatcase: userprofilepicture
  • UPPER-CASE-KEBAB-CASE: USER-PROFILE-PICTURE
  • Sentence case: User profile picture
  • Title Case: User Profile Picture
  • middot·case: user·profile·picture
  • dot.case: user.profile.picture
  • UPPER CASE: USER PROFILE PICTURE
  • lowercase: user profile picture

Update 12/06/20026

Checking that these changes to variable names and internal messaging do not clash with the GΓΆdel Machine.

Update 17/06/2026

The DevOps Engine (dvp) has unexpectedly proven to be critical to solving this puzzle. Mostly fixed last night. Watching the rather excellent live Google talk, Beyond the GPU: Maximising goodput with self-healing AI infrastructure, this morning has given me valuable insights into how to fix the remaining issues by reviewing Google HPC YAML files. 😎😎 Sometimes insights come from the strangest places.

Update 01/07/2026

The main work now is rapidly configuring Pipi for production and full autonomous automation. Using Google Search AI Mode (Gemini) and then Grammarly Pro makes the work easier and 100x faster.

  • I have decided to have Pipi re-render the many Ajabbi draft public websites with the new and missing developer information. (20K pages)
  • The website's .robot.txt file will then be unlocked to enable search engines.
  • The HTML will be updated to make it easier for AI to read.
  • This blog will be imported into Pipi, cleaned up, re-exported from Pipi, and published to Blogger via the API.
  • The new posts created in Pipi will return to A Sandy Beach to discuss something already built rather than being built.

Update 02/07/2026

The DevOps and IaC engines are getting rapid data model overhauls. The IaC engine is a great test for the variable names. I'm building a capability into Pipi to autonomously and automatically run OpenTofu and Ansible, initially targeting the Pipi Data Centre, then GCP and AWS for deployments. It's going very well and making rapid progress.

Update 05/07/2026

Pipi will initially run the open-source enterprise applications on Google Cloud Run and Google Cloud Storage (GCS). The code is complete and will be very low-cost to run, giving Ajabbi, a bootstrapping-purpose startup, a very long runway.

Update 18/07/2026

The job has now shifted to configuring, networking and deploying many physical servers. Installing software, including Pipi, labelling cables and rack gear, throwing out junk, tidying, etc., leaving nothing to chance. Shipping delays are holding up part deliveries.

Update 23/07/2026

On the basis of open collaboration and credits for experimentation, I was going to offer Google exclusive use of Pipi for a period (as a thank you) before Pipi open-source is donated to the Cloud Native Computing Foundation for all to use.

Make money to provide a service.

I'm getting exasperated with XWF. They are the external sales contractors to Google, and since 2021, they regularly contact me.

  • Selling GCP products (No need; I'm already convinced).
  • Acting as gatekeepers to any contact with Google Engineers to discuss novel integration options, which is the actual issue. How to combine Gemini (an LLM) and Pipi (non-LLM) to make something much better.
  • They are all very nice, but a complete waste of my time. No more XWF meetings, folks.

So, I have decided to target integration with OpenRouter (and its alternatives) instead of Gemini and open up the Pipi developer platform (it is big and coming 😎) to enable developers from Alibaba, Alice AI, Anthropic, AWS, Azure, ByteDance, DeepSeek, Google, IBM, Meta, Mistral, Moonshot AI, Naver, OpenAI, Oracle, Palantir, Sarvam AI, xAI, etc, and anyone else, to enable integrations that are optimal, 100% secure and vetted, with everything publicly verifiable.

Pipi closed-core will never be for sale; this year it's getting a non-profit foundation behind it, a bit like Patagonia. I'm open to all genuine offers of assistance, collaboration and experimentation with no strings attached. Contact me.

Don't send sales engineers

Send a senior, highly experienced engineer/architect/chief scientist who loves a big fat problem and has time for an open chat without a pitch or an agenda, and just see where it goes.

If you want to meet in person, expect to work collaboratively at a whiteboard or blackboard like a real mathematician. Plus coffee, of course. 😎 To see how this works, watch the seminars at the London Institute of Mathematical Sciences, or the recorded physics seminars at Perimeter.

Pipi is rooted in biology and the laws of physics, so you need a very solid background in advanced sciences (microbiology, biochemistry, mathematics, philosophy, particle physics, thermodynamics, complex adaptive systems, etc).

Please, no venture capitalists or private equity. You're wasting your time. Go find something else to plunder. Pipi is a gift to humanity.

Being very high-functioning Asperger's (autism) with hyperphantasia, plus multiple synesthesias, I think visually at lightning speed and output solutions as fast as I can draw. I love solving very hard problems that matter. I can only write very slowly with the help of assistive technology, so I prefer video meetings with good spoken English, slides and time for trading quick engineering drawings.

Pipi is designed to run massive enterprise systems for socially useful critical infrastructure on every platform in many languages and writing systems.

The intention is to make life better for all of humanity by destroying waste, failure and crippling bureaucracy in;

  • Health systems (hospitals)
  • Transport systems (rail, road, air, shipping)
  • Sewerage
  • Drinking water
  • Land drainage
  • Nature conservation
  • Built infrastructure
  • Electricity networks
  • GLAM (galleries, libraries, archives, museums)
  • Farming (agriculture, forestry, aquaculture, horticulture
  • etc
Infant mortality rates will be the KPI

Pipi makes these systems self-assembling, self-managing, resilient and adaptive. I had to solve hundreds of very big, hard, complex problems in parallel to make this work. Some of them were abandoned research by others, who couldn't make them work, so I solved them. The answers were there, hidden in plain sight. Invisible due to a lack of imagination or courage.

Easy for me, because I can do it visually in my mind, run simulations of thousands of components while sleeping, including the testing, then wake up and just build; it always works 100% (been doing it for decades). That's why no one else has cracked this problem. I can remember everything I have designed this way since age 4 in great detail. Curiosity-driven learning turns everything I read that's interesting into a moving 4d model in my mind; there are tens of thousands of these shimmering mental models, and they self-assemble when I shut my eyes to solve a problem. Each model grows in detail and size as I learn more. I can fly through the models, exploring and touching them. Really cool.

Ahaa moments most days

Some days, I wake up, and the insights and designs pour out of my head like a firehose, and I can barely keep up even after outputting 20+ drawings on A4 paper in a day. Now, there are many thousands of colour-coded drawings in ring binders.

I turn the growing backlog of these designs into data models, code, and documentation with references by giving simple, direct instructions to Google Search AI Mode (Gemini), which teaches me new skills and gives me a response to edit, test, correct, and use. I'm going 100x faster, like a bat out of hell, the equivalent of a crack team of pre-AI developers. 😎 And I'm getting a lot faster.

I rely 100% on intuition when surfing a sea of 4D visual mental models. I really don't understand how the rest of you can only think in words, because I can't.

I have been very lucky

My grandmother Bessie showered me with attention and love, gave me endless things to pull apart to see how they worked, and took me to meet very clever people in a small-minded backwater town.

Family holidays in wild New Zealand, next to rivers, beaches, forests and mountains, which ignited a lifelong obsession with the patterns of nature, the why.

My best friend right through school; he was the brightest kid in NZ.

My high school science teacher, Alan Morgan, let me play in the chemistry lab, doing experiments after school unsupervised for several years, and taught me the scientific method on my very last day at school, the most important thing I learned in 12 wasted years.

The wise old tradesmen, who took a skinny kid from sweeping the floor to being able to make anything, by learning on the job, trying hard, and having my butt kicked.

Nelson Mandela taught me to have the courage of my convictions and never give up.

The sculptor Neil Dawson and the set designer Tony Geddes taught me how to work authentically.

My blind friend Grant, who cut down bushes with a chainsaw and made and gave away $60 M, teaching me quiet courage and human decency.

The magnificent 50,000 working people of South Christchurch, who trusted me to lead a volunteer residents army doing recovery for 3 years, after the Christchurch Earthquake, teaching me humility and valuable leadership skills gained by trial and error in the moment.

My beloved Tracy, the bravest woman I have ever met, the only paraplegic to do the Coast-to-Coast Iron Man, who married a lost autistic male and made me a much better man. Her unwavering devotion, encouragement and loyalty made all this possible.

They all shaped me; I can't thank them enough. May their memories be a blessing.

The future is open, at the edge of chaos

Update 28/07/2026

Most of the equipment has arrived, and the small data centre setup is coming together. More deliveries later this week. It's already running a lot better and is much more productive.

Update 30/07/2026

My creative problem-solving and build process is evolving, and the loop is getting faster. Works every time.

  • Understanding the problem starts with reading or talking with someone, watching a YouTube talk, or listening to the radio.
  • Do lots of research, mostly reading books (days to months).
  • Drink plunger coffee.
  • Print off a paper(s) or article(s) and file it in an A4 3-hole ring-binder.
  • Drink Chamomile tea.
  • Daydream (solutions come within hours or months later).
  • Drink plunger coffee.
  • Draw the solution as many colour-coded A4 architecture drawing(s).
  • Drink more coffee (Cappuccino).
  • File the drawing(s) with the printed paper.
  • Use the drawings to prompt Google Search AI Mode (free) with detailed instructions on what to build.
  • Output teaches me, describes data model, code, documentation.
  • Print off and file with the rest.
  • Walk a dog. Plant a tree. Watch the sun rise.
  • Read the printed AI output, colour-code, add doodles, correct, test, edit names, build, use in Pipi, while listening to music. (I don't copy-paste. I manually type to copy, because it helps me learn and understand.)
  • Throw away all the paper except for the original research, which is moved from DevOps to the research library.
  • Pipi then generates self-documentation, including mermaid drawings.
  • Repeat.

Each loop cycle takes weeks to years. There are hundreds running in parallel at any one time. It's a pull system. I work on Pipi when something needs to be solved, and look up my library of solutions. Totally intuitive, just like an artist, not an engineer, and always fun like a kid playing.

These changes are made possible because of the detailed work done over the last few months removing barriers, including modifications to Pipi, reorganising space, equipment, and routines.

I think I have now solved all major problems to get Pipi 9 Core (Loki) running 24x7x52. Anything else that pops up can be solved along the way as part of maintenance.

Now to execute very fast.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library >
  • Home > Handbook > 

Last Updated

30/07/2026

No posts for a wee while

By: Mike Peters
On a Sandy Beach: 15/05/2026

Mike is the inventor and architect of Pipi and the founder of Ajabbi.

I was on a no-coding holiday for the last few weeks to clear my mind, and it has been great. I am back on the job today.

Suspended

Until the closed-source Pipi Core is back up and running 100% on autopilot, 10x faster, the following are suspended.

  • New posts "On a Sandy Beach
  • All newsletters, including the weekly Friday Report and the monthly Ajabbi Research Newsletter.
  • The fortnightly online Open R&D meeting.

Rapid refocus

  • A new developer area with five coding screens, designed to be more productive for hypervisual learners.
  • A better library has been set up for my A4 drawings in ring binders, the many reference books I use, and more bookshelves are on the way.
  • The server rack has been moved to a better location.
  • The light levels have been adjusted.
  • A big office tidy is almost done. An office-work-only desk has yet to be set up with a cat bed included.
  • A separate area with no screens for the happy cat, coffee, music, reading and drawing.

Less is more

Minimise screen time to be more productive at work. The new setup is also much less tiring.

Get the job done

The good thing is that, with a holiday and lots of drawing, I now have mental clarity about what needs fixing and how to fix it. Mainly, quite delicate changes here and there, organised into a list of steps. Now, I need to concentrate on one thing only: go as fast as possible, without meetings, post-deadlines, phone calls, or other distractions.

How

1. Use an AI workforce

Be the architect, and AI fills in the dots to make it happen.

Use Google Search AI mode (Gemini) to generate 99% of the code in one-page chunks (including references) to copy and paste, then manually change the variable names and SQL. Careful, test everything, resulting in 100x faster progress. Know how everything works and rapidly raise personal skill level.

2. Then build a cathedral

Make a wooden scale model of a cathedral for the builders. Google Search AI mode (Gemini) makes each brick, and Pipi Core assembles the bricks into floors, arches, walls, and vaults...

Speed is king

With the 100x coding productivity gains from Google Search AI mode (Gemini), plus the 10x10x10x speedup of Pipi Core currently underway over the next few months, what previously took a year will be done in hours and better.

Phase transitions

Once these initial migration issues from laptop to server are resolved, further transitions can be anticipated as the number of engines rapidly increases beyond 20. Increasing the number of engines slowly changes the whole system's behaviour from deterministic to probabilistic and adaptive.

Here is a partial list of transitions expected as the number of engines increases from 0 to 200. The actual numbers are a bit of a guess.

  • 20 engines enable Pipi 9 Core in a simple, deterministic structure.
  • 40 engines enable a workspace with a UI for administering Pipi Core.
  • 60 engines enable self-generation of user documentation.
  • 80 engines enable REPL and IAC (infrastructure-as-code).
  • 100 engines enable Workspaces for different user accounts.
  • Different Pipi 9 editions are made with the same engines, which recombine differently in response to the external environment.
  • And so on until...
  • 200 engines self-organise into a multi-layered complex fluid structure with probabilistic behaviour and emergent properties, as engines also act as agents.
  • 200+ engines enable Pipi 10 to interact with externally cloud-hosted LLMs, combining the very different strengths of both.
By , 2 comments:
Tags , , , , , , , , , , , , , , , ,

Live 4K video of Earth and space: 24/7 Livestream of Earth by Sen’s 4K video cameras on the ISS

Mike's Notes

Just beautiful.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library >
  • Home > Handbook > 

Last Updated

25/04/2026

Live 4K video of Earth and space: 24/7 Livestream of Earth by Sen’s 4K video cameras on the ISS

By: Sen
Sen: 25/04/2026

Sen’s vision is to democratise space through video to inform, educate, inspire, and benefit all of humanity. Sen’s mission is to stream real-time video from space to billions of people, gathering news and information about Earth and space and making the data universally accessible and useful.

Watch Earth Live and in 4K from Sen's video cameras on the International Space Station, downlinked via NASA. This is the world’s first continuous 4K livestream from space, empowering you to see our planet like astronauts do.

By , No comments:
Tags , ,

How to Accelerate Protein Structure Prediction at Proteome-Scale

Mike's Notes

Impressive and socially useful. The original article has many links.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > NVIDEA Developer
  • Home > Handbook > 

Last Updated

19/04/2026

How to Accelerate Protein Structure Prediction at Proteome-Scale

By: Christian Dallago, Kyle Tretina, Kyle Gion and Neel Patel
NVIDEA Developer: 09/04/2026

Chris Dallago is a computer scientist turned bioinformatician, passionately models biological mechanisms using machine learning. He's advanced bio-sequence representation learning, contributing to its establishment, notably in transformer models. Chris is dedicated to solving scarce data problems, such as designing proteins for therapeutic and industrial applications.

Kyle Tretina is a product marketing leader at NVIDIA, focused on advancing AI for digital biology and drug discovery. He drives the strategy and storytelling behind BioNeMo and our work with BioPharma, shaping how next-generation foundation models and GPU-accelerated microservices transform molecular and protein design. With a PhD in molecular microbiology and immunology, Kyle bridges science and strategy, translating breakthroughs in AI, chemistry, and biology into platforms that accelerate discovery for researchers, startups, and pharmaceutical companies worldwide.

Kyle Gion is a product manager for Research at NVIDIA, where he translates R&D in digital biology and molecular science into impactful products. He focuses on guiding research that applies computational biology, computational chemistry, and AI to life sciences, drawing on experience that spans both building scientific software and developing cystic fibrosis therapies. Kyle earned his bachelor's and master's degrees in Chemical Engineering from Brown University.

Neel Patel is a drug discovery scientist at NVIDIA, focusing on cheminformatics and computational structural biology. Before joining NVIDIA, Neel was a computational chemist in big pharma, where he worked on structure-based drug design. He holds a Ph.D. from the University of Southern California. He lives in San Diego with his family and enjoys hiking and traveling.

Proteins rarely function in isolation as individual monomers. Most biological processes are governed by proteins interacting with other proteins, forming protein complexes whose structures are described in the hierarchy of protein structure as the quaternary representation. 

This represents one level of complexity up from tertiary representations, the 3D structure of monomers, which are commonly known since the emergence of AlphaFold2 and the creation of the Protein Data Bank.

Structural information for the vast majority of complexes remains unavailable. While the AlphaFold Protein Structure Database (AFDB), jointly developed by Google DeepMind and EMBL’s European Bioinformatics Institute (EMBL-EBI), transformed access to monomeric protein structures, interaction-aware structural biology at the proteome scale has remained a bottleneck with unique challenges:

  • Massive combinatorial interaction space
  • High computational cost for multiple sequence alignment (MSA) generation and protein folding
  • Inference scaling across millions of complexes
  • Confidence calibration and benchmarking
  • Dataset consistency and biological interpretability

In recent work, we extended the AFDB with large-scale predictions of homomeric protein complexes generated by a high-throughput pipeline based on AlphaFold-Multimer—made possible by NVIDIA accelerated computing. Additionally, we predicted heteromeric complexes to compare the accuracy of different complex prediction modalities.

In particular, for the predictions of these datasets, we leveraged kernel-level accelerations from MMseqs2-GPU for MSA generation, and NVIDIA TensorRT and NVIDIA cuEquivariance for deep-learning-based protein folding. We then mapped the workload to HPC-scale inference by maximizing the utilization of all available GPUs, including scale-out to multiple clusters.

This blog describes the major principles we adopted to increase protein folding throughput, from adopting libraries and SDKs to optimizations to reduce the computational complexity of the workload. These principles can help you set up a similar pipeline yourself by borrowing from the techniques we used to create this new dataset.

So, if you are a:

  • Computational biologist scaling structure prediction pipelines
  • AI researcher training generative protein models
  • HPC engineer optimizing GPU workloads
  • Bioinformatician team building structural resources

You will learn how to:

  • Design a proteome-scale complex prediction strategy
  • Separate MSA generation from structure inference for efficiency
  • Scale AlphaFold-Multimer workflows across GPU clusters

Prerequisites

  • Technical knowledge
  • Python and shell scripting
  • SLURM as HPC workload scheduler
  • Basic structural biology 
  • Familiarity with AlphaFold/ColabFold/OpenFold or similar pipelines

Infrastructure

We describe scaling on a multi-GPU and multi-node NVIDIA DGX H100 Superpod cluster

This cluster includes high-speed storage to store MSAs and intermediate outputs

Software

  • Access to MMseqs2-GPU
  • Familiarity with TensorRT

If not using a model with integrated cuEquivariance, knowledge about triangular attention and multiplication operations 

Procedure/Steps

1. Define the dataset you’d like to compute

Begin by defining the scope of prediction. Because predicting protein complexes can become a combinatorial problem, it’s useful to understand what may be most interesting. In some cases, if your proteomes are small enough, an all-against-all (dimeric) complex prediction might be tractable; however, this could change if you want to predict large datasets of proteomes.

Here’s how we decided to go about it:

  • Homomeric complexes: We selected all proteomes represented in the AFDB and sorted them by perceived importance (e.g., proteomes of human concern or commonly accessed). This allowed us to rank proteomes for computation in a particular order, making execution more manageable.
  • Heteromeric complexes: This is where things can get complicated, fast. For our heteromeric runs, we decided to focus on complexes originating from several reference proteomes and proteomes included in the WHO list of important proteomes. As there’s an intractable number of combinations of complexes that can be derived from these proteomes, for our runs, we focused on dimers (complexes of two proteins), within the same proteome (no inter-proteome complexes) that had “physical” interaction evidence in STRING. As we sought coverage, we decided to consider all interactions reported in STRING for these proteomes, rather than further filtering. Evidence in the literature suggests that filtering for STRING scores >700 can further reduce the number of inputs while increasing the likelihood of well-predicted complexes.

2. Decoupling MSA generation from structure prediction

MSA generation and structure inference are both compute-intensive but scale differently, as we recently presented in a white paper. We thus approached these computations as separate steps and implemented separate SLURM pipelines. In general, for optimal use of a node, we set up MSA generation and structure prediction this way.

MSA generation

We generated MSAs using colabfold_search with the MMseqs2-GPU backend. While MMSeqs2-GPU scales across GPUs on a node natively, we chose to spawn one MMseqs2-GPU server process per GPU on a node for easier process management. In colabfold_search, the GPUs are only used for the ungappedfilter stages and not the subsequent alignment stages (which are multithreaded CPU processes).

Therefore, we can stack colabfold_search calls and start the next one once the GPU is no longer used by the previous one, by monitoring the colabfold_search output, to reduce GPU idle time.

Although this approach oversubscribes CPU resources, in practice, we found that on a DGX H100 node, up to 25% of the overall increase in throughput can be achieved with three staggered colabfold_search processes, at the expense of slower processing of individual input chunks. 

On determining reasonable input chunk sizes, there are two factors to consider. Smaller chunk sizes result in more chunks, which means more per-process overheads, such as database loading, which can take a couple of minutes each, even on fast storage. (Pre-staging the databases on the fastest storage available, such as the on-node SSD, helps with throughput as well.) On the other hand, larger chunks take more time to finish. On a SLURM cluster with a job time limit, this results in more unfinished chunks.

The sweet spot will depend on the cluster configuration, but for our DGX H100 node with a 4-hour wall time limit, the chunk size of 300 sequences seemed to work well with the staggering colabfold_search approach.

Structure prediction

In order to increase structure prediction throughput, we leveraged both optimizations in data handling for JAX-based folding through ColabFold, as well as accelerated tooling developed at NVIDIA, including TensorRT, and cuEquivariance for OpenFold-based folding.

Deep learning inference parameters

First, we selected inference parameters that struck a good balance between accuracy and speed. Protein inference setup for all deep learning inference pipelines (ColabFold and OpenFold), thus utilized:

  • Weights: 1x weights from AlphaFold Multimer (model_1_multimer_v3)
  • Four recycles (with early stopping)
  • No relaxation
  • MSAs: frozen MSAs generated through ColabFold-search (using MMseqs2-GPU), as described above

Accuracy validation

  Homodimer PDB set (125 proteins)
Model High Medium Accept Incorr Usable DockQ
DockQ >0.8 >0.6 >0.3 >0      
ColabFold 52 37 12 21 89 (72.95%) 0.637
OpenFold with TensorRT and cuEquivariance 53 39 10 20 92 (75.41%) 0.647

Table 1. A comparison of interface accuracy between ColabFold and OpenFold (accelerated by TensorRT and cuEquivariance) across a benchmark set of 125 homodimer proteins.

As we used different inference pipelines, we performed accuracy validation using a curated benchmark set of 125 X-ray resolved PDB homodimers released after AlphaFold2 was introduced, thus minimizing the potential for information leakage.

Predicted complexes for each deep learning implementation were compared against experimental reference structures using DockQ, which evaluates interface accuracy via the fraction of native contacts (Fnat), fraction of non-native contacts (Fnonnat), interface RMSD (iRMS), and ligand RMSD after receptor alignment (LRMS), and assigns standard CAPRI classifications of high, medium, acceptable, or incorrect.

Across the PDB homodimer benchmark, OpenFold accelerated through TensorRT and cuEquivariance reproduces ColabFold interface accuracy, achieving a similar fraction of “high” scoring predictions and comparable mean DockQ scores. This indicates that the accelerated implementations preserve interface-level structural accuracy relative to the ColabFold baseline.

MSA preparation and sequence packing

For ColabFold-based homodimer inferences, higher throughput can be achieved by packing homodimers of equal length into a batch for processing, sorted by their MSA depth in descending order. This reduces the number of JAX recompilations, thereby increasing end-to-end throughput. This trick, however, does not work when processing heterodimers, because the lengths of the individual chains differ.

For OpenFold, whether for homodimers or heterodimers, this packing strategy is not needed, as the method doesn’t require re-compilation. However, given a dependency between sequence length and execution time, reserving longer sequences for individual jobs may be beneficial if operating with specific SLURM runtimes. To further optimize the process, input featurizations (CPU-bound) were performed for the next input query alongside the inference step for the current query (GPU-bound).

Additionally, OpenFold’s throughput was enhanced through the integration of the NVIDIA cuEquivariance library and NVIDIA TensorRT SDK. These modular libraries and SDKs can be leveraged to accelerate operations common in protein structure AI and general inference AI workloads, respectively. We previously described how TensorRT can be leveraged to accelerate OpenFold inference.

3. Optimize GPU utilization with SLURM

As alluded to in the previous section, depending on the available hardware, you can increase throughput by “packing” GPUs and nodes. SLURM is a great orchestrator, and we divided the inference workflows in SLURM scripts to:

  • Pack multiple predictions per node
  • Match GPU memory to sequence length
  • Reduce idle time between jobs
  • Separate short vs long sequence queues

Our workload was mapped to a H100 DGX Superpod HPC system. We could thus deploy inference across NVIDIA H100 GPUs on multi-node clusters, leveraging exclusive execution on a single node, and packing each GPU with as many processes as saturated the GPU utilization for both MSA processing and deep learning inference.

Helpful tips:

  • Group jobs by total residue length
  • Monitor GPU memory fragmentation
  • Use asynchronous I/O to avoid disk bottlenecks

4. Making quality predictions accessible to the world

In partnership with EMBL-EBI, the Steineggerlab at Seoul National University, and Google DeepMind, we explored complex structure prediction analysis. We highlight that predicting these biological systems remains challenging. Unlike protein monomer prediction, where predicted Local Distance Difference Test (pLDDT) can inform overall prediction quality, yielding a balanced amount of plausible predictions, in the complex scenario, assessing interface plausibility is much harder. This has to do with the fact that assessing complexes involves global and per-chain confidence metrics, as well as local confidence metrics at the interface.

Simply put, is the interface between two monomers plausible, and is it predicted in the right pocket? These questions are much harder to answer than more “local” questions about monomer likelihood, given the very limited data available. Therefore, we make available a set of high-confidence structures through the AlphaFold Database, thereby enabling, for the first time, exploration of protein complexes. We intend to refine our approach further and expand the universe of available protein complexes in the AlphaFold Database.

Getting started

Proteome-scale quaternary structure prediction requires more than just running AlphaFold-Multimer at scale. Success depends on:

  • Evidence-driven interaction selection
  • Decoupled and optimized compute workflows
  • GPU-aware job orchestration
  • Confidence calibration and validation
  • Dataset health monitoring

By combining STRING-guided selection, MMseqs2-GPU acceleration, and NVIDIA H100-powered multimer inference, this work extends AFDB into a unified, interaction-aware structural resource.

This infrastructure enables:

  • Variant interpretation at interfaces
  • Systems-level structural biology
  • Drug target validation
  • Generative protein design benchmarking

Resources

Read more about the project here: https://research.nvidia.com/labs/dbr/assets/data/manuscripts/afdb.pdf 

Accelerated libraries and SDKs are available here:

  • MMseqs2-GPU
  • NVIDIA cuEquivariance
  • NVIDIA TensorRT

If you wish to deploy MSA search and protein folding easily, you can get accelerated inference pipelines through NVIDIA’s Inference Microservices (NIMs):

  • MSA Search NIM 
  • OpenFold2 NIM

The predictions from this effort are available through https://alphafold.com

By , No comments:
Tags , ,

Is Particle Physics Dead, Dying, or Just Hard?

Mike's Notes

This Quanta Magazine article about the current state of Particle Physics got me thinking.

  • Particle physicists are very good at statistical physics and maths
  • They have to be very bright and well-trained
  • They like hard problems to solve
  • There are unemployed particle physicists
  • Ajabbi Research will need people in the future. Hmmm

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Quanta Magazine
  • Home > Handbook > 

Last Updated

20/03/2026

Is Particle Physics Dead, Dying, or Just Hard?

By: Natalie Wolchover
Quanta Magazine: 26/01/2026

Natalie Wolchover is a columnist for Quanta Magazine, where she covered the physical sciences for more than a decade. Her writing has been featured in The Best American Science and Nature Writing, The Best American Magazine Writing, and The Best Writing on Mathematics, and has won several awards, including the 2022 Pulitzer Prize for Explanatory Reporting, the 2016 Evert Clark/Seth Payne Award, and the American Institute of Physics’ 2017 Science Communication Award. She was lead editor for the National Magazine Award–winning special issue, "The Unraveling of Space-Time." Her first book, The Question to Which the Universe Is the Answer, is scheduled for publication in 2027.

Columnist Natalie Wolchover checks in with particle physicists more than a decade after the field entered a profound crisis.

In July 2012, physicists at the Large Hadron Collider (LHC) in Europe triumphantly announced the discovery of the Higgs boson, the long-sought linchpin of the subatomic world. Interacting with Higgs bosons imbues other elementary particles with mass, making them slow down enough to assemble into atoms, which then clump together to make everything else.

A couple of months later, I took a job as the first staff reporter at the nascent science magazine that would become Quanta. Turns out I was starting on the physics beat just as the drama was picking up.

The drama wasn’t about the Higgs particle; by the time it materialized at the LHC there was already little doubt about its existence. The Higgs was the last piece of the Standard Model of particle physics, the 1970s-era set of equations governing the 25 known elementary particles and their interactions.

More striking was what did not emerge from the data.

Physicists had spent billions of euros building the 27-kilometer supercollider not only to confirm the Standard Model but also to supersede it by uncovering components of a more complete theory of nature. The Standard Model doesn’t include particles that could comprise dark matter, for instance. It doesn’t explain why matter dominates over antimatter in the universe, or why the Big Bang happened in the first place. Then there’s the inexplicably enormous disparity between the Higgs boson’s mass (which sets the physical scale of atoms) and the far higher mass-energy scale associated with quantum gravity, known as the Planck scale. The chasm between physical scales — atoms are vastly larger than the Planck scale — seems unstable and unnatural. In 1981, the great theorist Edward Witten thought of a solution(opens a new tab) for this “hierarchy problem”: Balance would be restored by the existence of additional elementary particles only slightly heavier than the Higgs boson. The LHC’s collisions should have been energetic enough to conjure them.

But when protons raced both ways around the tunnel and crashed head-on, spraying debris into surrounding detectors, only the 25 particles of the Standard Model were observed. Nothing else showed up.

In philosophy, “qualia” refers to the subjective qualities of our experience: what it’s like for Alice to see blue or for Bob to feel delighted. Qualia are “the ways things seem to us,” as the late philosopher Daniel Dennett put it. In these essays, our columnists follow their curiosity, and explore important but not necessarily answerable scientific questions.

The absence of any “new physics” — particles or forces beyond the known ones — fomented a crisis. “Of course, it is disappointing,” the particle physicist Mikhail Shifman told me that fall of 2012. “We’re not gods. We’re not prophets. In the absence of some guidance from experimental data, how do you guess something about nature?”

Once the standard reasoning about the hierarchy problem had been shown to be wrong, there was no telling where new physics might be found. It could easily lie beyond the reach of experiments. The particle physicist Adam Falkowski predicted to me at the time that, without a way to search for heavier particles, the field would undergo a slow decay: “The number of jobs in particle physics will steadily decrease, and particle physicists will die out naturally.”

The crisis and its fallout made for years of interesting reporting, but sure enough, the frequency of news stories related to particle physics diminished. I fell out of touch with sources. More than 13 years on, in this first column for Qualia, a new series of essays in Quanta Magazine, I’m taking stock. Is particle physics dying, as Falkowski predicted? Can new physics still be found? What’s the future for particle physicists? Will artificial intelligence help? How much hope is left in the search for answers to the many remaining mysteries of the universe?

Some particle physicists act as if there’s no crisis at all. The LHC is still running and will for at least another decade, and its operators are finding new sources of enthusiasm.

In the last couple of years, data handling at the collider has improved with the use of AI. Pattern recognizers can sort through the outgoing debris of proton collisions and classify collision events more accurately than human-made algorithms can. This helps the physicists to more accurately measure the “scattering amplitude,” essentially the probability that different particle interactions will occur. For instance, AI systems can determine more precisely how many top quarks arise in the aftermath of collisions versus the number of bottom quarks. Any statistical deviations from the predictions of the Standard Model could signify the involvement of unknown elementary particles.


A proton-proton collision documented by the Compact Muon Solenoid at CERN in 2012 shows evidence of the decay of the Higgs boson.

CMS Collaboration; Mc Cauley, Thomas

Novel particles as hefty as Higgs bosons would not be so subtle; they would have shown up already as pronounced bumps on data plots. But as Matt Strassler, a particle physicist affiliated with Harvard University, explained to me, the traces of lighter novel particles could still lie in so-called hidden valleys in the data. “There’s a huge amount of unexplored territory there,” he said. There might exist, for instance, an unstable type of dark matter particle that leaves its mark by occasionally arising and immediately decaying into an excessive number of muon-antimuon pairs. Detecting such an excess would point indirectly to the unstable particle’s existence. “For people who thought all the new physics is at high energies — they’re very disappointed right now,” Strassler said. “I don’t share that view. There are many opportunities for nature to provide clues at low energies.”

So far, though, no such indirect evidence of new physics has been detected. The more accurate the statistics have become at the LHC, the better they match the Standard Model. Michelangelo Mangano, a particle physicist at CERN, the laboratory that houses the LHC, said the collider today is like a tool for exploring the Standard Model’s predictions, and he considers this exploration worthwhile because not all consequences of the equations are easy to calculate. The search for new physics beyond the Standard Model is ongoing, Mangano said, but “the fact that it’s not giving positive results does not mean we are stuck, dead, or wasting our time.”

These questions are so fundamental that of course it’s worth nailing down every amplitude and checking every hidden valley, since we have the tool for the job. But for hunters of new physics, does the game end there?

The community wants to go bigger. CERN physicists want to build a Future Circular Collider, tripling the circumference of the LHC with a 91-kilometer tunnel beneath the Franco-Swiss border, to both probe higher energies and look for subtler signals. This FCC would initially collide electrons, which, unlike protons, are themselves elementary particles, with no substructure. Their clean collisions would allow more precise measurements of scattering amplitudes, making the FCC ultrasensitive to indirect signs of new physics. By the end of the century, the mega-collider would be upgraded to collide protons, as the LHC does now. Proton collisions are messier, but at the FCC they would achieve unprecedented energies — about seven times higher than the LHC can currently muster — so they have a chance, however slim, of revealing heavy particles beyond the LHC’s reach. (In theory, particle masses could range up to a million billion times greater than what the LHC energy scale can produce directly, so there’s no reason to expect them around the next bend.)

We’re not gods. We’re not prophets. In the absence of some guidance from experimental data, how do you guess something about nature? - Mikhail Shifman

As of now, the FCC’s fate is unknown; formal approval and funding commitments by member countries won’t come before 2028.

Meanwhile, U.S. particle physicists are aiming to complement the European strategy by constructing a brand-new type of machine: a muon collider. Muons are elementary like electrons, but they’re 200 times heavier, so their collisions would be both clean and energetic (albeit not reaching the collision energies of the LHC). Both the selling point and the challenge of this newfangled type of machine is that it will require major technical innovations (with all the spin-off potential that can bring), because muons are highly unstable. They must be accelerated and collided mere microseconds after they’re created.

Demonstrating the technology and then constructing the collider would take roughly 30 years, and that’s with federal funding. “We have to figure out how to do it in between 10 and 20 billion [dollars],” said Maria Spiropulu, a physics professor at the California Institute of Technology and co-chair of the committee behind a national report endorsing a muon collider program(opens a new tab) that came out in June 2025. Over the coming years, the Department of Energy will weigh whether to fund the proposal rather than competing science projects. What hurts its case is the lack of a “discovery guarantee,” which the LHC had with the Higgs boson.


Scientists and technicians inspected and upgraded systems at the Large Hadron Collider during the Long Shutdown 2, which began in 2018.

Maximilien Brice/CERN

Then again, as the mathematical physicist Peter Woit mused on his blog(opens a new tab), “Perhaps in our new world order where everything is controlled by trillionaire tech bros, the financing won’t be a problem.”

Deliberations about a Chinese supercollider have come to naught, I’m told. Instead, China has decided to pursue a “super-tau-charm facility”: a lower-energy particle scattering experiment that would cost mere hundreds of millions of dollars instead of tens of billions. The facility will produce a lot of tau particles and charm quarks, partly to study whether taus ever shape-shift into muons or electrons. This kind of switching isn’t predicted by the Standard Model, but it does happen in some theoretical extensions of it.

Okay, we might as well check. We’re desperate for new physics, and the price is good. But by definition it’s very difficult to know which shots in the dark are worth taking.

Adam Falkowski, who sounded the death knell for particle physics back in 2012, used to be known for the sharp commentary he supplied on his blog RΓ©sonaances(opens a new tab). But the Paris-based particle physicist hasn’t posted anything since 2022. He said that’s partly because he’s been tied up with fatherhood and partly because there hasn’t been much to say.

When we caught up on a video call, Falkowski told me, “I am very skeptical about future colliders. For me it’s very difficult to get excited about it.” He sees momentum behind CERN’s FCC campaign, but personally he worries about the huge costs and timescales, and the fact that “there are absolutely no hints that something is there within the reach of the next collider.”

For his part, Falkowski has turned to the theoretical study of scattering amplitudes, a growing research area focused on the geometric patterns underlying particle interaction statistics, patterns that could point toward a truer perspective on the quantum world. The field seeks to reformulate the equations of particle physics in a different mathematical language in hopes that this language might extend to quantum gravity. “There is a very vibrant program in trying to understand the structure of the physical theories,” Falkowski said. “The hope is that with the help of machine learning, that there can be very fast progress in the coming years. I think that’s where the best things have happened.”

But amplitudeology, as this field is known, is abstract — it’s no atom-smashing experiment. Falkowski said he does think experimental particle physics is dying. He has watched talented postdocs switch to other research areas or take data science jobs. “I’m not sure they are getting the best of the best as they used to,” he said, “because the prospects of returns are so distant. If you want to change the world now, you will do AI; you will do something different from particle physics.”


The ALICE (A Large Ion Collider Experiment) detector at the Large Hadron Collider was designed to study quark-gluon plasma.

CERN, Julien Marius Ordan/Science Source

This brain drain appears to be real. I spoke to Jared Kaplan, co-founder of Anthropic, the company behind the chatbot Claude. He was a physicist the last time we spoke. As a grad student at Harvard in the 2000s, he worked with the renowned theorist Nima Arkani-Hamed to open up the new directions in amplitude research that are being actively pursued today. But Kaplan left the field in 2019. “I started working on AI because it seemed plausible to me that … AI was going to make progress faster than almost any field in science historically,” he said. AI would be “the most important thing to happen while we’re alive, maybe one of the most important things to happen in the history of science. And so it seemed obvious that I should work on it.”

As for the future of particle physics, AI makes worrying about it now rather pointless, in Kaplan’s view. “I think that it’s kind of irrelevant what we plan on a 10-year timescale, because if we’re building a collider in 10 years, AI will be building the collider; humans won’t be building it. I would give like a 50% chance that in two or three years, theoretical physicists will mostly be replaced with AI. Brilliant people like Nima Arkani-Hamed or Ed Witten, AI will be generating papers that are as good as their papers pretty autonomously. … So planning beyond this couple-year timescale isn’t really something I think about very much.”

Cari Cesarotti, a postdoctoral fellow in the theory group at CERN, is skeptical about that future. She notices chatbots’ mistakes, and how they’ve become too much of a crutch for physics students. “AI is making people worse at physics,” she said. “What we need is humans to read textbooks and sit down and think of new solutions to the hierarchy problem.”

Cesarotti was a high school junior when the Higgs boson was discovered. She grew up near Fermilab, the U.S. national lab in Illinois that houses the Tevatron, which was the world’s highest-energy particle collider before the LHC. (The top quark was discovered there in 1995.) This proximity taught her that a particle physicist was a thing you could be. Later, it turned out to be her thing. “What are the fundamental building blocks of the universe — those were the questions that I was most interested in knowing the answer to,” she told me. “But what people said was, ‘Particle physics is dead. Don’t do this.’”

It may have been a fair warning; Cesarotti has yet to land a permanent job as a rising particle physicist. The subfield has continued to shrink, she and others said, as faculty hiring committees and grad students go in other directions. “Definitely all this rhetoric that there was nothing to be found and you should give up on it — people listened,” she said. “And of course that means there are fewer people. It becomes a self-fulfilling prophecy. If you’re pushing all these talented people out of trying to solve these problems into a field that it’s easier to make an impact on, then you’re setting yourself up for failure.”

Cesarotti echoed a sentiment I’d heard from others, which sounds correct to me as well: “Particle physics isn’t dead; it’s just hard.” It’s hard to know what to think about or look for. But the most devoted particle physicists are thinking and looking all the same.

“It was easy for 125 years,” Strassler said. “One thing led to the next. That lucky century has, for now, at least in the medium term, come to an end. That could change tomorrow, or next century, or who knows.”

A hint of a new lightweight particle could, in theory, show up at the LHC, or in some other experiment. Strassler is particularly excited about the study of radioactive thorium-229 decay, which could reveal variations in the fundamental constants. I’m slightly partial to experiments looking for “axions,” dark matter candidates that are so lightweight that they can act a little like light itself.

On the theory side, an obvious solution to the hierarchy problem could drop naturally out of the geometry behind scattering amplitudes. Or, if Kaplan is right, AI systems might someday suggest powerful new ideas for how the 25 particles of the Standard Model fit into a more comprehensive pattern — a possibility I didn’t foresee back when the crisis began.

Clearly, further progress toward the truth remains possible in particle physics. But there’s no discovery guarantee. I’ve had more than 13 years to think about it, and it remains a disturbing prospect: All the empirical clues we can glean about nature’s fundamental laws and building blocks might already be in hand. The universe may plan on keeping the rest of its secrets.

By , No comments:
Tags , ,

These New AI Models Are Trained on Physics, Not Words, and They’re Driving Discovery

Mike's Notes

A fantastic use of AI. My instinct is to incorporate fluid-like systems into a future Pipi. That will require a real data centre.

Resources

References

  • Reference

Repository

  • Home > Ajabbi Research > Library > Subscriptions > Simons Foundation
  • Home > Handbook > 

Last Updated

19/03/2026

These New AI Models Are Trained on Physics, Not Words, and They’re Driving Discovery

By: Elizabeth Fernandez
Fatiron Institute: 09/12/2025

Elizabeth Fernandez is a science writer specializing in science and society, science and philosophy, astronomy, physics, and geology.

While popular AI models such as ChatGPT are trained on language or photographs, new models created by researchers at the Flatiron Institute and other members of the Polymathic AI collaboration are trained using real scientific datasets. The models are already leveraging the knowledge they learn from one field to address seemingly completely different problems in another.

The Walrus AI model simulates fluid motion.

Walrus/Polymathic AI

While most AI models — including ChatGPT — are trained on text and images, a multidisciplinary team of scientists has something different in mind: AI trained on physics.

Recently, members of the Polymathic AI collaboration presented two new AI models trained using real scientific datasets to tackle problems in astronomy and fluidlike systems.

The models — called Walrus and AION-1 — are unique in that they can apply the knowledge they gain from one class of physical systems to seemingly completely different problems. For instance, Walrus can tackle systems ranging from exploding stars to Wi-Fi signals to the movement of bacteria.

That cross-disciplinary skill set is particularly exciting because it can accelerate scientific discovery and give researchers a leg up when faced with small samples or budgets, says Walrus lead developer Michael McCabe, a research scientist at Polymathic AI.

“Maybe you have new physics in your scenario that your field isn’t used to handling. Maybe you’re using experimental data, and you’re not quite sure what class it fits into. Maybe you’re just not a machine-learning researcher and just can’t burn the time working through all the possible models that might fit your scenario,” McCabe explains. “Our hope is that training on these broader classes makes something that is both easier to use and has a better chance of generalizing for those users, as the ‘new’ physics to them might be something another field has been handling for a while.”

Using cross-disciplinary models can also improve predictions when data is sparse or when studying rare events, says Liam Parker, a Ph.D. student at the University of California, Berkeley, and a lead researcher developing for AION-1.

The Polymathic AI team recently announced Walrus in a preprint on arXiv.org and presented AION-1 on Friday, December 5, at the NeurIPS conference in San Diego.

Walrus and AION-1 are ‘foundational models,’ meaning they’re trained on colossal sets of training data from different research areas or experiments. That’s unlike most AI models in science, which are trained with a particular subfield or problem in mind. Rather than learning the ins and outs of a particular situation or starting from a set of fundamental equations, foundational models instead learn the basis, or foundation, of the physical processes at work. Since these physical processes are universal, the knowledge that the AI learns can be applied to various fields or problems that share the same underlying physical principles. Foundational models have a host of benefits — from speeding up computations to performing well in low-data regimes to finding physics shared across different fields.

AION-1 is a foundational model for astronomy. It is trained on data from astronomical surveys that are already massive in their own right: the Legacy Survey, the Hyper Suprime-Cam (HSC), the Sloan Digital Sky Survey (SDSS), the Dark Energy Spectroscopic Instrument (DESI) and Gaia. All in all, that’s more than 200 million observations of stars, quasars and galaxies totaling around 100 terabytes of data. AION-1 uses images, spectra and a variety of other measurements to learn as much as it can about astronomical objects. Then, when a scientist obtains a low-resolution image of a galaxy, for example, AION-1 can extract more information about it, learned from the physics of millions of other galaxies.

Walrus’ domain is fluids and fluidlike systems. Walrus utilizes the Well — a massive dataset compiled by the Polymathic AI team. The Well’s data encompasses 19 different scenarios and 63 different fields in fluid dynamics. All in all, it contains 15 terabytes of data describing parameters such as density, velocity and pressure in physical systems as wide-ranging as merging neutron stars, acoustic waves and shifting layers in Earth’s atmosphere.

Such foundational models can be powerful. AION-1 and Walrus can utilize physics seen in a different case and apply it to learn about something new. It is similar to our senses. “Multiple senses together — rather than one at a time — gives you a fuller understanding of an experience,” the AION-1 team explained in a blog post about the project. “Over time, your brain learns associations between how things look, taste and smell, so if one sense is unavailable, you can often infer the missing information from the others.”

Then, when a scientist is performing a new experiment or observation, they have a starting point — a map of how physics behaves in other similar situations. “It’s like seeing many, many humans,” says Shirley Ho, Polymathic AI’s principal investigator and an astrophysicist and machine learning expert. Ho is a senior research scientist at the Flatiron Institute and a professor at New York University. When “you meet a new friend, because you’ve met so many people before now, you are able to map in your head … what this human is going to be like compared to all your friends before,” she says.

Foundational models make scientists’ lives easier by streamlining data processing. Scientists will no longer have to create a new framework from scratch for every project or task; instead, they can start with an already trained AI to use as a foundation. “I think our vision for some of this foundation model is that it enables anyone to start from a really powerful embedding of the data that they’re interested in … and still achieve state-of-the-art accuracy without having to build this whole pipeline from scratch,” says AION-1 lead researcher Parker.

Their goal is to make tools that scientists can use in their day-to-day research. “We want to bring all this AI intelligence” to the scientists who need it, Ho says.

Other Highlights From the NeurIPS 2025 Conference

CosmoBench: CosmoBench is a multiview, multiscale, multitask cosmology benchmark for geometric deep learning. Curated from the state-of-the-art cosmological simulations, CosmoBench is the largest benchmark of its kind, with over 34,000 point clouds and 25,000 directed trees. CosmoBench features challenging evaluation tasks from cosmology and diverse baselines, including cosmological methods, simple linear models and graph neural networks. This presentation will show how CosmoBench is pushing the frontiers of cosmology and geometric deep learning.

Lost in Latent Space: Physicists model and predict the behavior of physical systems using their understanding of the laws of physics. However, these calculations require significant computing power. Flatiron Institute scientists and other members of the Polymathic AI collaboration studied whether a less taxing form of computing can still yield accurate results. Known as ‘latent diffusion modeling,’ this computational model utilizes artificial intelligence to generate high-quality images at a lower computational cost while accurately capturing physical behavior.

Neurons as Detectors of Coherent Sets in Sensory Dynamics: Our perception of touch, taste, sight and pain is mediated by neurons that carry signals from peripheral receptors to the brain. This work shows that these neurons can be understood as detecting ‘coherent sets’ within the sensory stream — groups of stimulus trajectories that evolve together over time and therefore share a common past or a common future. By distinguishing these coherent sets, some neurons predominantly encode what has just occurred, while others reliably signal what is likely to happen next. Traditional classifications of sensory neurons can thus be reinterpreted as reflecting a division between past-focused and future-predictive processing. Understanding how the nervous system separates and transforms sensory input in this way may offer new routes for treating mental illness and may also guide the development of biologically inspired artificial intelligence.

Predicting Partially Observable Dynamical Systems: Scientists can predict the motion of a falling object or the evolution of fluids using deterministic models that compute a single future outcome from past observations. But this approach breaks down for physical systems where much of the state is hidden. A prominent example is the sun: We can observe the activity on its surface, but the processes deep inside remain largely invisible. Without access to those internal conditions, there isn’t enough information to forecast a single ‘correct’ future. Researchers at the Flatiron Institute, together with collaborators in the Polymathic AI project, have developed a probabilistic approach that can infer these hidden solar processes. By incorporating information from the distant past into a diffusion-based generative model, their method produces an ensemble of plausible futures, offering a clearer understanding of how past sunspot activity shapes its future evolution.

By , No comments:
Tags , , , ,
Subscribe to: Posts (Atom)