Ferenc Huszár Portfolio
Professor of Machine Learning at Cambridge University and founder of Reasonable
Professor of Machine Learning at the University of Cambridge, on extended leave to build Reasonable, his startup focused on superhuman programming LLMs; previously Senior Machine Learning Researcher at Twitter after it acquired Magic Pony Technology; his "inFERENCe" blog runs 12 pages back to 2012, with technical posts on GANs, causal inference, and discrete diffusion, plus a January 2026 retrospective that grades his own 2016 post "Deep Learning is Easy - Learn Something Harder."
Location
Remote-first, in the most literal sense — no idea where.
Companies
Credentials
- PhD in Bayesian machine learning, University of Cambridge Engineering Department
Skills
Notable
- Co-author of "Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network" (SRGAN, CVPR 2017; 17k+ Google Scholar citations)
- Co-author of the sub-pixel convolution super-resolution paper (CVPR 2016; ~10k Google Scholar citations)
- Senior Machine Learning Researcher at Twitter; helped set up Twitter's ML Ethics, Transparency and Accountability (META) team, per his Cambridge profile
What makes it work
A breakdown of the choices that make this portfolio stand out.
Theoretical rigor applied to trend-driven topics
Posts on diffusion models or LLM behavior are grounded in probability theory and statistics rather than restating whatever is trending, which is what makes the analysis hold up years after publication.
Long-view retrospectives, not just reactions to new papers
The "10 Years On" post revisits his own 2016 post and marks each prediction as a hit or a miss, with the misses (architecture tweaks, scaling) flagged as plainly as the hits — a rarer form of intellectual honesty than only ever writing about the newest result.
A single-author blog sustained across more than a decade
The archive runs to twelve pages, from 2012 posts on social influence to 2026 essays on LLMs and software, evidence of a genuinely sustained personal research practice rather than a blog active for one hype cycle.
What AI Researchers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Ground trend-topic posts in the underlying theory (probability, statistics, optimization) rather than restating the trend — it's what keeps writing relevant after the trend passes.
Revisit your own older claims publicly against new evidence — it's a stronger credibility signal than only ever writing about what's new.
Sustain a single-author blog over years rather than a burst tied to one hype cycle — the archive itself becomes the credential.
Projects
Twelve pages of posts reach back to 2012, and the newest ones include Ferenc Huszár marking his own old predictions with clown emoji, a genre of peer review most researchers leave to other people.
SRGAN: Photo-Realistic Single Image Super-Resolution
A GAN for image super-resolution from his Magic Pony Technology period. Its perceptual loss is, in his words, 'a change that has been universally adopted in almost all follow-on work'.
Industry research at a startup that turned into a field-standard technique.
Came out of a startup later acquired by Twitter
Loss-function change rather than an architecture tweak
Revisited in his 2026 retrospective
Deep Learning is Powerful Because It Makes Hard Things Easy: Reflections 10 Years On
A January 2026 post grading his own 2016 post 'Deep Learning is Easy - Learn Something Harder', item by item, with clown and target emoji marking misses and hits.
Publicly auditing your own past claims is a strong credibility signal.
Revisits a post that reached the top of Hacker News
Explicit hit/miss format
Ends by asking who is wrong today
Causal inference blog series
A multi-part series starting with an introduction to causal inference and do-calculus and running through causal diagrams, Markov factorization, and structural equation models (part 4, June 2021).
Explains a mathematically dense topic to ML practitioners, one part at a time.
Four-part series
Grounded in probability theory
Part of a 12-page archive
More AI Researcher portfolios
Ready to build your portfolio?
Follow the AI Researcher roadmap — skills, projects, and timeline to get hired.