NeurIPS 2025 Tavel Blog

Flight to San Diego

I stood in another terminal for my flight to San Diego. Having gone through two flights the previous month for my trip to Washington, D.C., I was well prepared for the flight to San Diego. I left Orlando International Airport (MCO) and landed the comparatively small San Diego International Airport (SAN) at about 9PM local time. From there, I proceeded to my Comfort Inn Hotel in the historic Gaslamp Quarter.

An interesting note about SAN is that the airport is built on a small island in the harbor very near the heart of the city. Further, the airport is saddled with a single, short runway, and there is no room on that island for expansion. Despite that limitation, the airport enjoys "international" status and connects mainland America to the discontinuous states of Alaska and Hawaii and nations on the Pacific Rim including Canada, Japan, Mexico, and Panama. To keep up with demand, planes came and went every two minutes from the airport for those various destinations; and they could be seen and, in places, heard flying very low over the city every day of the conference.

The Gaslamp Quarter

The Gaslamp Quarter is small segment of San Diego near downtown and across the street and a set of train tracks from the San Diego Convention Center that hosted NeurIPS 2025. It was there that I stayed in a Comfort Inn on the corner of 7th and G St. There was no shortage of 19th century architecture to enjoy in the Gaslamp Quarter, and I took time out of my morning walks, on different routes each day, to the convention center to photograph some of them.

Along with the tens of thousands of other conference goers that week in December, I would pass through that district of the city and head to the famed San Diego Convention Center in the morning. At night, we would all descend on the Gasplamp District to socialize and have dinner.

Food in the Gaslamp Quarter

When I checked into my hotel, it was nearly 10pm Monday night. Much like my stay in Washington D.C. a month prior, I was arriving late enough that I could expect most places to be closed. I spoke with the concierge at the front desk about whether there would be any places open to eat that night, and he thankfully pointed me to a high-quality fast-food establishment called The Melt.

Following his advice, I went to the restaurant and ordered a light dinner: a grilled cheese sandwich with a side order of sweet potato fries. The sandwich was somewhat sweet and perfectly melted, so that the cheese was stringy and molten but not liquid. The potatoes were well salted, although some retained a slight crispness. I thoroughly enjoyed it but forgot to take a picture.

Instead, here is a picture of another meal I ate there a couple nights later. This double burger was more visually impressive both in its size and how the meat patties were melted together by the cheese. In this image, we also see the generous portioning of the sweet potato fries, which cost less than $5 dollars. I reckon that almost a whole sweet potato's worth of material is in the basket I received, and the cardstock basket itself is nearly too small to hold the load.

Ghirardelli is a famous, high-end chocolatier that I have tasted several times in my life, and they have stores in many cities across the nation, although they aren't as numerous as many ice cream parlor chains that one might imagine. I compare Ghirardelli to creameries since they serve ice cream confections at their brick-and-mortar locations in addition to selling their chocolate, which is made in store. I had tasted their ice cream at to their Disney Springs location in Orlando, so I was pleasantly surprised to discover their famous San Diego location next door to The Melt. One night, mid-conference, I went in for a hot fudge sundae.

Very common in San Diego's Gaslamp Quarter are the numerous taquerias, usually featuring counter service and outside seating. I ate at several of them during my time there, including Tacos el Franc and Taco Centro. At The Melt, Ghirardelli, the taquerias, and other locations, I met numerous people from the conference. Until the weekend of the 6th and 7th, it seemed like virtually every person on the street in that part of the city was associated with the conference. Crowds of researchers crowded every eatery in the evenings, especially the bars, and I met many of them as they waited for their orders near the counters of taquerias and fast-food restaurants. Our numbers and the crowds we managed to form in the Gaslamp Quater's eateries are an unusual but strong testament to the size of the machine learning research community.

The Forecasting Company Mixer

The Forecasting Company (TFC) ran a mixer on Sunday, the last day of the conference, at a bar named The Blind Burro which served us a taco buffet with beans, chicken and beef available as the protein; guacamole and salsa to spread over the warm corn tortillas, and toppings of onion, tomato, and lettuce. It took place in the evening after the conference reached its conclusion that afternoon and was attached to the Recent Advances in Time Series Foundation Models (BERT2S) that was held that afternoon. It was in the room next door to the LLM Evaluation workshop that I presented my OPTiCAL poster in that afternoon, so I stopped by to see what was being presented there and to refresh the acquaintance I had already made with Lele Cao and his team at King's Stockholm AI Lab. In the workshop, I learned about the latest advances in time series foundation models and the role they play in time series forecasting tasks.

When I had traversed the city and made my way to The Blind Burro, I was excited to continue discussing those topics and to get to know both the researchers from BERT2S and the mixer hosts from TFC better. In the shadow of Petco Park, home stadium of the San Diego Padres MLB team, I stood with about a dozen other workshop attendees and waited for the event to begin. TFC co-founder Joachim Fainberg was checking event registrations and letting people into the event through a false, open doorway in the middle of the bar. Awkwardly, there was no wall to one side of this archway at the end of one wall, only a table that would have made navigating around it a bit difficult.

I was one of the first people through the door. Fainberg handed me a sticker that said TFC which identified me as one of the guests at their party; I attached it to the paper card on the lanyard I had been wearing all that week; and I waited for the meal to be served while dozens of other conference goers flooded the rear portion of the bar behind that doorway. I sat with two men that were conversing in Scandinavian accents at a high-top table near a window that looked out at the scaffolded facade of Petco Park, and a waitress came around with an appetizer of cheese flautas served with salsa verde for dipping. The three of us shared the flautas, and I enjoyed the somewhat minty salsa as we waited to for dinner to begin in earnest. When Lele Cao appeared from the doorway and sat with the three of us, addressing my two new acquaintances jovially, I was surprised. Without knowing it, I had been sitting with two Swedish employees of King's AI Lab in Stockholm who were attending the conference with Lele Cao.

After dinner began and our plates were made, I spent most of the dinner conversing with them, speaking about my research and looking back on the research that I had seen at the conference. A woman named Maddy Wang Renström who is the director of AI/ML at King. I learned quite a bit about their lives and work in Stockholm, and we discussed at length the experience and logistics of flying half-way around the world from Sweeden to California and how my journey was comparatively easy, short, and cheap. The dinner stretched on for a couple hours like that.

Then Fainberg approached the center of the room and told everyone to gather on one side of it, where there were booths already filled with people. We were told to gather ourselves for a group photo featuring the BERT2S attendees who made it to the mixer. Everyone gathered on that side of the room and organized themselves. I stood in the middle row and center frame, but before the photo could be taken the co-founder took his chance to explain the event and his business to everyone.

Fainberg said that TFC was an apply named start up time series forecasting company that specialized in serving models for forecasting tasks, hence their interest in BERT2S. His company was hiring at all levels, he said, and he would be available after the photo op to discuss the opportunities awaiting those of us who were interested.

With that, the camera snapped, and my night at the TFC mixer at NeurIPS 2025 was immortalized by the company's photographer. Imagine my pleasure at seeing the image appear on LinkedIn later. I have embedded the LinkedIn post in this webpage. All credit and thanks to The Forecasting Company for making the mixer possible and to its co-founder Geoffrey Négiar for sharing the photo with everyone.

Special Thanks

Additionally, special thanks to NeurIPS Foundation, the San Diego Convention Center, and the city of San Diego for hosting the many thousands of researchers who attended NeurIPS 2025. Without their meticulous planning and dedication to the event, we would not have been able to gather and share the latest developments in machine learning research from labs around the world. I hope to attend it again next year or in the years following and know that I will take everything I learned there with me as my blossoming career continues.

For more information about our research, return to our homepage: ufdatastudio.com.