About
I love solving hard business problems. I like to say, "There are an infinite number of…
Activity
2K followers
-
Daniel Marthaler shared thisI can say that having worked with Steven in the past, this is one of the best opportunities I could imagine for ML engineers to pursue. His mentorship and attention to detail in project scope and implementation is the chance of a lifetime! None of you should pass this up!Daniel Marthaler shared thisWe're hiring on Meta's Ads Ranking team (Sequence Learning & User Modeling)! 🚀 We're looking for ML engineers at all levels: from new grads to Staff. The pitch: Bring LLM-style scaling to recommender systems. We train frontier-scale sequence models and ship them into production ranking that serves billions of ad requests per day at low latency. What we're working on: 📈 Scaling user sequence models to ever-larger model sizes and sequence lengths 🧠 Integrating semantic content understanding and world knowledge into recommender systems 🤖 Building generative approaches to recommendation This is one of the few places where you can do cutting-edge sequence modeling research and see it run in production at massive scale. The models you build directly move Meta's ads ranking. 🎯 Interested? DM me… happy to chat! Location: Sunnyvale, CA; Bellevue, WA; New York, NY #Hiring #MachineLearning #RecSys #SequenceModeling #LLM
-
Daniel Marthaler reposted thisDaniel Marthaler reposted thisI’m angry, and it's hard to put this into words. Sexual demeaning and child sexual abuse material (CSAM) are not inevitable consequences of technology. It's a deliberate product strategy by X and Elon Musk. The environment that allows this to flourish has been around for some time. AI amplifies these existing flaws in the system and exacerbates the harms. This is a well-known, and pretty obvious if you think about it for a bit. We've had a disregard for safety on social media, general lack of attention and legal enforcement for sexual crimes against women and children, the absence of business ethics, a lack of product sense, a subjugation of empathy to misogynistic status incentives for tech folks. And now a leader who is taking advantage of that with no regards for the harms caused. If you post on X, please stop. If you have power or a position where you can keep these harms from being normalized, please do it.
-
Daniel Marthaler shared thisJohn is an amazing leader and would be amazing to work with! You should apply if you are qualified!!Daniel Marthaler shared thisSuperhuman is developing a world-class AI-native productivity suite, complemented by a proactive AI assistant designed to work wherever you do. We are seeking exceptional growth and product data scientists for various mid- to senior-level positions. If you have extensive expertise in experimentation, product analytics, or marketing measurement and are interested in shaping the future with us, please send me a resume via direct message or apply directly through the link provided. https://lnkd.in/gMYyzhJU
-
Daniel Marthaler shared thisI read the responses to this joke and wonder, "why don't you people write tests?" 🤔
-
Daniel Marthaler shared thisRachel is Awesome! and awesome to work with!! If you are qualified, you should apply!Daniel Marthaler shared thisThe data team at Midi is growing! We're hiring for 3 data roles at Midi Health across different experience levels. Come join the fun and help women get access to the healthcare they deserve 🚀
-
Daniel Marthaler shared thisPatric is awesome! His group is doing some of the coolest stuff out there. If you are qualified, you should apply!Daniel Marthaler shared thisHello all! I'm excited to share that I am hiring an L6 Research Scientist for my team! I am looking for an experienced candidate to research, develop, and iterate on LLM-as-a-Judge prototypes and offline LLM evaluation systems to improve Netflix’s understanding of our members. NOTE: These posts can generate a lot of interest, so if you think this role might be a good fit for you, please reach out to my recruiter Anthony Rivera! Otherwise I might accidentally miss your message. In this role, I am looking someone to help drive how Netflix thinks about our members and systems. Some examples of questions we are thinking about are: 1. How should we think about integrating member outcomes (such as what people play, what they like) along with human feedback in LLM judges? 2. How should we evaluate the quality of a Netflix homepage that consists of images, text, rows, etc.? 3. How can LLM judges be used to bridge the gap between survey results and behavioral data? Job is remote friendly. Again, if you are interested, please reach out to Anthony Rivera! Posting attached below! https://lnkd.in/eETmHEvAResearch Scientist (L6) - LLM-Driven Product Understanding | USA - Remote | NetflixResearch Scientist (L6) - LLM-Driven Product Understanding | USA - Remote | Netflix
-
Daniel Marthaler shared thisThis is a fantastic opportunity to see what "real data" looks like in practice, and to acquire amazing experience with one of the best colleagues I know! You (yes you!) should apply!
-
Daniel Marthaler shared thisCome for the great work, stay for the algebraic topology discussions! Bryan is a great leader and amazing colleague. If you are qualified, you should apply!!Daniel Marthaler shared thisI am back on the hunt for AI Engineers! For those of you who have been reaching out about when I will have more headcount, the moment has arrived! Theory Ventures is hiring two SWEs: first is AI eng focused and second is Infra focused. Roles are currently open, please DM me. A great profile is Mid-career, high-ambition, SFBA located (1day in office required). References on me from my previous teams are available upon request, but as a snapshot, here's what some of them have said: - "Bryan is kinda funny" - "I like that Bryan doesn't try to tell me how to implement things" - "Bryan wears cool clothes" But more seriously, let's cook.
-
Daniel Marthaler reposted thisDaniel Marthaler reposted thisThe layoffs of the last few weeks in the US public sector have impacted a number of colleagues/collaborators in the environmental sciences. Some of those collaborators have reached out (not just to me, but to lots of my colleagues in the AI space as well) to ask for advice on building up their AI skills before re-entering the workforce. So, a few of us who have benefited from a zillion amazing collaborations with the US public sector are organizing a two-week remote workshop in May that will teach computer vision and machine learning methods to former US public sector environmental scientists. Please share with anyone who might be interested, including potential participants (i.e., "students") and potential volunteers/instructors. https://lnkd.in/ggzKgVsV
-
Daniel Marthaler liked thisDaniel Marthaler liked thisEveryone in AI now talks about world models, the thing that LLMs are supposed to lack and that might be the next AI revolution. But what are world models? Is this a new idea? Are there several kinds of world models? While new architectures are being developed (e.g. JEPA, DINO-WM, etc), aren’t there ways to train LLMs to actually acquire world models? These questions are addressed in the new blog post of the Flowers AI & CogSci team Centre Inria de l'université de Bordeaux, written by the amazing Clément ROMAC! After linking with the old idea of model-based reinforcement learning, the post starts discussing world models as most often conceptualized in AI: the ability to predict the future given the current state, history and action. In machine learning, training for this ability is often seen as running special loss functions over special architectures (e.g. JEPA) given a dataset of trajectories of the world. Then, the post discusses various other ways to think of world models as human cognition uses them, and as studied in cognitive sciences, e.g. models for explanation, intuitive theories, formal models, etc. Finally comes the core of the post about how human world models are acquired through self-exploration and self-experimentation of the world, often driven by curiosity, and how it’s possible to implement this in generative AI models… even in LLMs that were supposed to be incapable to acquire world models. To this end, Clement describes one recent project of the team: WorldLLM, a technique transforming LLMs into curiosity-driven agents that can spontaneously explore their environment, forming intuitive human-readable hypotheses about the world dynamics, then planning for experiments to test them, and iteratively refining them and testing new ones as new data are incrementally acquired. That was a project with Guillaume Levy Cédric Colas Thomas CARTA The post concludes by positioning WorldLLM within the larger research program of autotelic AI. Read here: https://lnkd.in/efqAymHi
-
Daniel Marthaler liked thisDaniel Marthaler liked thisI'm hiring a Platform Engineer to join my team at Stitch Fix! This role is on the Dev Platform team where we build the internal platform-as-a-service and foundational infrastructure empowering our engineers to safely and efficiently deploy and operate their services. We're an AWS shop, consolidating on EKS, and use Terraform, Pulumi, Ruby on Rails, Golang, and more. I'm looking for someone late junior - early mid in their career that has foundational infrastructure and development experience and is interested in applying that to improve capabilities for other engineers. We're remote-first and distributed across the US. Happy to answer questions about the role or the team, and please spread the word if you know someone who might be interested!
-
Daniel Marthaler liked thisDaniel Marthaler liked thisLast week, I had the opportunity to visit Café Momentum's headquarters and local Dallas (aka Café Momentum Dallas) operations and meet many of my new teammates. As someone who has been motivated throughout my career by economic opportunity and social justice, I am so excited to be part of an organization whose ambitions to change how our society approaches youth justice are only eclipsed by the courage, resilience, and dreams of the youth we serve. Café Momentum's mission is to transform young lives by equipping justice-involved youth with life skills, education, and employment opportunities to help them achieve their full potential. We do this by providing youth ages 15–19 with 12-month paid internships in one of our restaurants located in Dallas, Atlanta, and Pittsburgh — with Denver coming soon — coupled with wraparound support including mental health services and 24/7 case management. Meeting with teammates, including our Founder Chad Houser, Jessica Blais Metcalf, Porshia Haymon, Ph.D., Olivia Cole, and many others, put me in touch first-hand with the blood pumping through the veins of this organization. Meeting with the youth in the program and an alum who now represents the organization as an Ambassador was incredibly inspirational and reminded me of the profound responsibility I and my colleagues have in stewarding the organization and representing it to the community of would-be funders, strategic partners, and compatriots in the youth justice and workforce development space. I also had the chance to visit the construction site of our new Dallas Flagship — a facility that will allow us to train our team, build community, and share the Momentum Model with the world. The sign on the fence says it all: "We're building more than a restaurant." Indeed we are. If you'd like to learn more about how Café Momentum is helping kids get onto a better path — reducing recidivism, advancing educational attainment, and cultivating life and professional skills — or about what we've learned serving the unique workforce development needs of justice-involved youth, hit me up. Let's chat! And if you happen to live in or be passing through a city where we have a restaurant, dine out with us. Just be sure to wear stretchy pants because the food is that good.
-
Daniel Marthaler liked thisDaniel Marthaler liked thisIn 1610, Johannes Kepler reflected on a radical possibility in a letter to Galileo Galilei: what if every fixed star were itself a Sun? If so, perhaps those distant suns might also have worlds around them. Worlds that would remain hidden until “someone equipped for marvelously refined observations” could detect them. Four centuries later, that challenge has become one of the great engineering and scientific frontiers of our time: not just inferring that exoplanets exist, but directly imaging them. At Stanford University Space Rendezvous Laboratory, we have approached this problem from the perspective of distributed space systems: how can a telescope and a free-flying occulter, or starshade, work together across enormous separations to suppress the light of a host star and reveal the faint signal of planets and exozodiacal dust nearby? This idea motivated the Miniaturized Distributed Occulter/Telescope mission concept, mDOT, a small-satellite starshade architecture designed to demonstrate high-contrast imaging through precision formation flying, navigation, and control. More recently, Zahra Ahmed’s research has pushed this work toward the next frontier: applying modern simulation, post-processing, and machine learning to starshade images for exoplanet detection and characterization. Her work includes convolutional neural networks for detecting exoplanets in starshade imagery, as well as the latest study on ultraviolet starshade capabilities for exo-Earth imaging in support of future Habitable Worlds Observatory concepts. What I find beautiful is the continuity of the question. Giordano Bruno and Johannes Kepler could only speculate that other suns might have other worlds. Today, we are designing the distributed spacecraft, optical systems, and AI-enabled data pipelines that may allow us to see and characterize them directly. Check here for our publications on the topic: https://lnkd.in/e4tps8mt Thanks Shane Lowe for finding the citation of Kepler’s communication to Galileo: “The passage appears in Johannes Kepler’s 1610 response to Galileo’s Sidereus Nuncius, titled Dissertatio cum Nuncio Sidereo, commonly translated by Edward Rosen as Kepler’s Conversation with Galileo’s Sidereal Messenger. #Exoplanets #Starshade #DirectImaging #FormationFlying #DistributedSpaceSystems #SpaceAutonomy #HabitableWorldsObservatory #Stanford #SpaceRendezvousLab
-
Daniel Marthaler liked thisDaniel Marthaler liked thisas a dyslexic high school kid, i had a rough time with algebra. i loved the subject, but i was slow and error-prone. i wanted software to replace paper and pencil while still allowing me to explore math. it turns out that Geoffrey Irving wrote a paper describing exactly this software and a technique to make it work. https://lnkd.in/gMC9QKhs i have just released the first draft of algebranch. it is open source, runs locally, and requires no login. the source code is available at: https://lnkd.in/gVNTiQKJ you can use it now at: algebranch.org/ i would love to get this in front of math educators and students like me. all feedback is encouraged, and there is a handy in-app feedback button. to be clear, algebranch does not solve equations; it only supports algebraic manipulation. it is old school - so no in-app AI.
-
Daniel Marthaler liked thisDaniel Marthaler liked thisWe're #hiring a Senior Manager, Data Science leading our Foundational Models here at Stitch Fix. This role is for someone who gets excited about the part of machine learning that becomes real infrastructure for a business: the scoring, ranking, and matching systems that connect clients to inventory and power personalization across the experience. What makes this one especially interesting is the mix. The work is deeply technical, but it also sits close to real product decisions, real client outcomes, and real business tradeoffs. It’s not research at a distance. It’s building the core models that help make personalized shopping actually work. If you love leading strong teams, shaping strategy, and turning AI/ML into something durable, useful, and measurable, I’d love for you to take a look.
-
Daniel Marthaler liked thisDaniel Marthaler liked thisTwo work updates for me: (1) Winding down VarietyIQ, and (2) Joining Literati as Head of Data Science. ▪️VarietyIQ was awesome. We built great things, learned a lot, and had fun – but it’s time to close that chapter. I'm very grateful to everyone who supported our journey. ▪️I’ve been working part time with Literati for the past year. The team, mission, and day-to-day work are all great. The company is also at a clear inflection point and I’m excited to help build from here. Let's go team! PS: I’m still blogging on operations, data-science, and beyond – now at jslandy.com
-
Daniel Marthaler liked thisDaniel Marthaler liked thisExciting news from Stitch Fix today! Now, clients can create Vision images of themselves in recommended outfits. I created this one (with a skirt I love and needed ideas for how to style). Read more about it...and try it. It's fun. https://lnkd.in/gWZKAjx5
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
-
Arizona State University
-
-
Dissertation Title: On two problems in Dynamical Systems: 1. Optimal Supply Chain Management with Re-entrant Semiconductor Factories.
2. Bootstrapping Errors for Lyapunov Exponents from Time Series Embeddings
Publications
-
Metamaterials design using gradient-free numerical optimization
Journal of Applied Physics
We apply numerical optimization methods in combination with full-field electromagnetic simulations to tailor the broadband spectral response of gold and silver split-ring resonator metamaterials. A derivative-free, nonlinear mesh adaptive search technique was used to drive finite-difference time-domain simulations. This algorithm allows the designer to independently vary the spectral position of the two resonant peaks and their relative reflection amplitudes throughout a wide range of the near…
We apply numerical optimization methods in combination with full-field electromagnetic simulations to tailor the broadband spectral response of gold and silver split-ring resonator metamaterials. A derivative-free, nonlinear mesh adaptive search technique was used to drive finite-difference time-domain simulations. This algorithm allows the designer to independently vary the spectral position of the two resonant peaks and their relative reflection amplitudes throughout a wide range of the near infrared. An application of this method is then shown to design split-ring resonator "notch filters," with narrow pass bands at 1310, 1550, and 1800 nm which have an similar to 45% change in reflectivity at the pass band and corresponding linewidths of similar to 90 meV.
Other authors -
Projects
-
Gaussian Process Python Toolbox
- Present
An extension of Carl Rasmussen's GPML Matlab toolbox in python. In partnership with Marion Neumann in the CS Department at Washington University in St. Louis, Kristian Kersting at Technical University of Dortmund and Shan Huang at SAP Berlin.
Other creators -
Languages
-
Spanish (working knowledge)
-
Other similar profiles
Explore more posts
-
James Brand
Wrote a second blog post! This time it's about using LLMs for imputing missing data. I've seen a few papers about imputation with LLMs, but most seem to either train custom imputation transformers or horserace flagship models against standard methods. Mert Demirer and I recently played with an alternative idea which instead uses LLMs as part of an ensemble imputation approach, relying on their "knowledge" of the world to provide additional prediction signal. I haven't seen a paper about it yet, so we thought it'd be fun to make a short post: https://lnkd.in/gMzC6iaP
8 Comments -
DataMListic
Metropolis-Hastings is one of those algorithms that shows up everywhere in Bayesian statistics, yet often feels unintuitive at first. This video explains how the Metropolis-Hastings algorithm works, why acceptance probabilities matter, and how MCMC enables sampling from complex distributions used in modern machine learning and data science. Watch the full video here: https://lnkd.in/dMAJqRSD
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top contentOthers named Daniel Marthaler in United States
6 others named Daniel Marthaler in United States are on LinkedIn
See others named Daniel Marthaler