Yann LeCun: Very cool work from UCL+FAIR-London on neural architectures for reasoning that learn to generate and select rules.
3 replies, 316 likes
Pasquale Minervini: Conditional Theorem Provers are scalable neuro-symbolic reasoning models that learn to recursively select and generate rules on-the-fly conditioned on the goal via gradient-based optimisation! To appear at #ICML2020, Arxiv https://arxiv.org/abs/2007.06477 Slide http://neuralnoise.com/icml20_talk.pdf 1/N https://t.co/QbqiTaXH24
6 replies, 315 likes
Tim Rocktäschel: The @icmlconf 2020 talk on "Learning Reasoning Strategies in End-to-End Differentiable Proving" by @PMinervini is now online: https://slideslive.com/38928094/learning-reasoning-strategies-in-endtoend-differentiable-proving
1 replies, 94 likes
Tim Rocktäschel: Excited to share our @icmlconf 2020 paper on more scalable end-to-end differentiable proving, enabling longer reasoning paths than our previous NTPs or GNTPs. SotA on CLUTTR raw graphs for link prediction. Great work by @PMinervini! w/ @riedelcastro, Pontus Stenetorp & @egrefen
0 replies, 55 likes
Pasquale Minervini: Feel free to join our paper discussion at https://icml.cc/virtual/2020/poster/6364, happening right now!!
0 replies, 23 likes
Edward Grefenstette: Happy to announce the release of @PMinervini's awesome work on further improving and scaling Neural Theorem Provers by jointly learning to propose rules to consider during the NTP proof process. New SOTA and better systematic generalization on CLUTRR. Upcoming at #ICML2020.
0 replies, 9 likes
Pasquale Minervini: @spolu @arvind_io hi! you may be interested in some work on neural theorem proving: https://arxiv.org/abs/1705.11040 (NeurIPS 2017), https://arxiv.org/abs/1912.10824 (AAAI 2020), https://arxiv.org/abs/2007.06477 (ICML 2020)
0 replies, 5 likes
HotComputerScience: Most popular computer science paper of the day:
"Learning Reasoning Strategies in End-to-End Differentiable Proving"
0 replies, 5 likes
Amir Saffari: Great paper by @PMinervini and colleagues on end-to-end differentiable methods for learning reasoning
0 replies, 2 likes
Pasquale Minervini: @arankomatsuzaki @riedelcastro @xwhan_ @sriniiyer88 @JefferyDuu @PSH_Lewis @WilliamWangNLP @YasharMehdad @scottyih Amazing work! It seems very related to https://arxiv.org/abs/2007.06477 - we take the goal and reformulate it in one or more sub-goals (using a distinct dense encoder for each of them), and match them against a set of facts via MIPS or reformulate it further. Love seeing this on HotpotQA!
0 replies, 1 likes
注目の最新arXiv【毎日更新】: 2020/07/13 投稿 2位
Learning Reasoning Strategies in End-to-End Differentiable Proving
5 Tweets 46 Retweets 175 Favorites
1 replies, 0 likes
Found on Sep 16 2020 at https://arxiv.org/pdf/2007.06477.pdf