Burst Mode: Professor Josh Bloom

Professor Josh Bloom in his office (Photo by Johnny Gan Chong)

September 3, 2024
Avi Rosenzweig

Professor Joshua Simon Bloom first joined the Astronomy Department in 2005. By then he had already bounced between the US and the UK, between the ivy-covered brickwork at Harvard and the sun-drenched courtyards at Caltech, and there would still be many career ricochets to come before he landed in the Chair’s desk at UC Berkeley. In 2020. Just as the pandemic lockdown was unfolding. Perfect. He ran the department during COVID the same way he aims to in all the projects he leads: by striving to provide the best tools and right support so that everyone involved can do their best work. This has been the animating force for Bloom all his life. Give bright people what they need to do their best and everything (else) will pretty much work out well. 

The initial stages of Bloom’s launch into astronomical research owe as much to luck as passion for science. During his undergrad in astronomy and physics at Harvard, he landed a summer internship at Los Alamos National Lab (LANL) working with Dr. Edward Fenimore, just when time domain astronomy was getting a boost from new technological capabilities – new satellites were spotting ever more high-energy explosions from outside our solar system, and computing was commanding an increasingly central role in making sense of the growing volumes of astronomical data. 

Artist's concept of a Gamma Ray Burst (GRB) (Credit: Nasa/Swift/Mary Pat Hrybyk-Keith and John Jones)

Dr. Fenimore was the leading astrophysicist in the race to understand the physics of gamma-ray bursts (GRBs), quick flashes of gamma and X-rays coming from seemingly random places in the sky. Much of what astronomers have their eye on are static phenomena like galaxies and nebulae that change very slowly, on the scale of millions to billions of years. Come back to the observatory tomorrow and those will be right where you left them. But Fenimore and his confrères were chasing something different: transient phenomena like supernovae and GRBs. Blink twice and you’ll miss them. To catch a better view of these you need instruments that get great action shots, or have a wide field of view, or that can record for a long time like a security camera. Ideally, all three. This is what Fenimore was putting together when Josh Bloom started to work with him at LANL. A mystery raged about the origin and distances to these events. He headed early editions of the Swift Observatory satellite that look at these transient phenomena. Not only can it find many more GRBs, “the spacecraft ‘swiftly’ (in less than approximately 90 seconds) and autonomously repoints itself to bring the burst location within the field of view of the instruments to observe the afterglow.” Gamma rays are like an electro-house bass music throb that could be coming from anywhere, but add in the other measurements and the origin can be pinpointed and studied further. Now time domain specialists could rake in data as fast as their computers allowed. Bloom published two papers as an undergrad with Fenimore on GRBs, and went on to graduate magna cum laude, class of ’96. (Fifteen years later, Bloom would write a comprehensive book on GRBs called “What Are Gamma-Ray Bursts”, published by Princeton University Press.)

While taking advantage of Harvard’s connection with Emmanuel College to do a Master’s in astrophysics at Cambridge University, Josh hit the jackpot again. He joined the group of Martin Rees, the éminence grise of British theoretical physics. Rees had recently predicted the existence of “afterglow” radiation following a GRB and so Bloom quickly got caught up in observing new GRB events with optical telescopes, hunting for this afterglow. Astronomers, including Bloom, had been theorizing that supernovae would give off an afterglow since the early 1990s, but the first confirmed sighting, detected by a new Italian-Dutch satellite, didn’t happen until the beginning of 1997, which happened to be when Bloom was crunching numbers at Cambridge. How lucky was that? The burst lasted for about 80 seconds, and skywatchers were able to localize it and get x-ray readings and other data really quickly because of the instruments that Fenimore’s group had been developing. It was coming from a supernova 8.1 billion lightyears away. This turned “what had been a search for needles in a haystack into shooting fish in a barrel”: just as Rees had predicted, there was optical, infrared, and even radio afterglow. The GRB field was transformed.

Spotting GRBs and catching their afterglow became the darling of the astronomy industry. Josh worked full time on GRBs as a graduate student at Caltech, finding evidence that the majority of GRBs come from exploding massive stars. He and a UC Santa Cruz professor wrote one of his highest cited articles, a review called “The Supernova—Gamma Ray Burst Connection”. In 1998 Bloom was awarded the Hertz Foundation Fellowship as his interests began to include instrumentation. Between 1997 and 2001 a pair of synced telescopes in Arizona and Chile cataloged over 300 million astronomical objects. The AZ scope was automated and retooled for a project Josh Bloom headed:  PAIRITEL. As a Junior Fellow at the Harvard Society of Fellows, Bloom put his energy into building an autonomous afterglow discovery engine using that mothballed telescope. His project became the first robotic telescope operating at infrared wavelengths and found the first contemporaneous infrared afterglow of a GRB just a few weeks before Bloom joined the astronomy faculty in Berkeley. PAIRITEL data eventually served as a basis for several PhD theses (on exoplanets and supernovae) and over 75 academic papers. 

Sculpting all this new data into shapes that would hold water, especially when it came from many different sources and needed to be ready fast, had Bloom writing up new software scripts and data analysis hacks himself. He realized that he had a knack for it. He jokes that astronomers have been responding to big data problems for hundreds of years by “hiring more graduate students to sit down and look at the images.” Bloom was open to trying out new computing methods just as he was new telescopes, and he had yet another fortunate encounter, this time with early versions of machine learning (ML) models. Astronomical images coming from lightyears away are noisy and hard to understand. But if an AI can learn to read handwriting, could it be used to spot a new event in the hazy sky? Can uncertainty quantification in one area be applied to another? You bet. Bloom partnered with computer science and statistics colleagues to start one of the earliest data science centers in the country, The Berkeley Institute for Data Science (BIDS) in 2009.  He offered Python programming bootcamps and graduate seminars that drew the largest enrollment in the department graduate course catalog. He learned that students were not getting the appropriate training for the research they expected to accomplish. They needed more than a user manual for a program that would be obsolete in a year; they needed to understand how deep learning works. He could create the workers he needed by giving them the right tools. He enjoyed tinkering under the hood of ML models enough to careen off in another career direction.

Professor Josh Bloom in his office (Photo by Johnny Gan Chong)

In 2012 Josh and some of his collaborators took the engine out of the ML model they had built to speed up astronomical analyses and founded a startup company that “helps business users regain control over complex decision making in the face of ever-growing and increasingly diverse data.” It did well, getting contracts with well known firms like Lyft and Expedia. Well enough that in 2016 they were acquired by GE. Josh Bloom went from an accomplished time domain astronomer to a vice president of GE’s Digital division. Ask him to tell you what it was like to be in board meetings with GE CEOs to hear about different kinds of transient phenomena.

When he re-aimed back to the Astronomy Department in the late 2010s, Bloom had now become an experienced research leader: expertise in the subject of time domain marvels and wonders, a skilled hand at using the edgiest data models, and experience managing radiant young scientists within a larger, always-in-flux, organization. 

How, now, to use the constraints of the pandemic lockdown to lean into potentially transformative research workflows? AI tools are already handy for speeding up tasks that staff must plod through. Bloom is looking for more – a change in the kinds of problems that surface, rather than just continuing down the same roads quicker. His group should take on “the novelty squared problem” of finding questions that will interest both the astronomy and the data science communities, problems whose solutions improve our grasp of astronomy and of data theory, plus perhaps clarify the foundations of physics. 

But why stop there? The management and blue-sky idea sides of Professor Bloom have him picturing a new way of organizing the data infrastructure of research programs everywhere into a new kind of sharing network, a kind of internet of data collections, that could nip the Large Language Model (LLM) data scarcity problem in the bud. 

Now is not the point at which we say “Slow down there a minute, Sport. Let’s keep the ball inside the service lines.” Bloom knows when to push the pace. He is picturing a multidisciplinary center to cross-train young investigators in AI and astronomy. The kind of institute that will hold symposia and hackathons. A place where wide surveys of the skies can be transformed from data points into universal stories. The moment is ripe for this kind of pas de deux. 

The moment is also ripe to bring a managerial perspective to big data – a moment to learn from big business. Bloom sounds like the founders here on the left coast when he preaches about the help we will be able to give to those with bright futures that need a friendly environment to bloom. Intellectual property laws and data categorization frameworks are fetters that keep our best from running with their ideas. If we can get lawyers and LLMs talking to each other the horizon will stretch so much farther. Bloom imagines a network of data source arrangements that constitute a “federated learning” system where each holder of data resources will be able to take what they need and give what they are able without compromising anyone’s privacy or proprietary interests. Intellectual work that comes out of “interesting collaborations” is what Professor Bloom now aims to enable.

The cycle of scientific progress still keeps Joshua Bloom swinging. He underlines that there is no clear start or finish to the theory/experiment/iterate circle. “You keep on going around.” He already has an eye on what’s just cresting the hill—maybe neural computation models will reveal new discoveries in astronomical data? So long as he can mentor rising scholars, collaborate with bright colleagues, and spread the stellar tools we have at hand, astronomy will continue to shine. 

Professor Josh Bloom at the Campbell Hall Rooftop Observatory (Photo by Johnny Gan Chong)

Topics

Faculty topic page, Mathematical & Physical Sciences topic page, Feature topic page