Imperial's ERC AToM project hiring 1 postdoc, 2 PhDs for adaptive tokenization research
PontiEdoardo · x · 2026-09-11
Edoardo Ponti at Imperial College is recruiting 1 postdoc and 2 fully funded PhD students for his ERC project AToM (Adaptive Tokenization and Memory), aiming to build models that learn to compress what they perceive and remember for permanent memory, longer horizons, and >10x speedups.
The core problem: internal representation length in dominant architectures is fixed upfront by tokenization, creating quadratic attention bottlenecks and KV-cache bandwidth costs. The lab has prototyped architectures that learn, end-to-end, to compress representation sequences:
- Up to 8x speedups with no accuracy loss
- Longer effective input/output horizons under a fixed budget
- Enables lifelong learning via sub-linearly growing memory, inference-time hyper-scaling, and richer multimodal world modeling
They've open-sourced retrofitted models (Qwen3-8B-DMS-8x with NVIDIA, Bolmo-7B with AI2). Postdoc deadline Sept 30; PhD rolling; B200s and H200s available plus funding for 2 conferences/year.
More from Research
- Tao and Fields Medalists' two objections to AI in math, and why they're weak — RexDouglass · 2026-09-12
- Conjectures launches Bittensor bounties paying TAO for cracking math problems open 30-80 years, judged by machine — markjeffrey · 2026-09-12
- FADA (CoRL 2026) open-sourced: humanoid robots adapt to new conditions from 2 minutes of experience — GuanyaShi · 2026-09-12
- Intel's silicon photonics couplers hit 1-1.5 dB IL, with visible epoxy delamination flaws — jwt0625 · 2026-09-12
- CPO paper criticized for vague DLW-to-PIC coupling description: 'such as TCB' — jwt0625 · 2026-09-12
- Fly connectome trained to play a Flappy Bird–style game — TinfoilTricorn · 2026-09-12