You can add location information to your Tweets, such as your city or precise location, from the web and via third-party applications. You always have the option to delete your Tweet location history. Learn more
Bongard analogy challenge uses 6 positive and negative examples to convey a concept (e.g. convex). Original one from 1960s ~ few hundred hand-engineered problem instances. Instead, we use LOGO to programmatically generate concepts at scale => no data scarcity for #DeepLearning
Human evaluation accuracy on our benchmark was as high as 99% for careful evaluators. Yet, all current meta/few-shot/self-supervised methods < 70%. Even the best #DeepLearning methods are not able to learn underlying simple concepts at human level.
We carefully designed benchmark to not reward overfitting by #DeepLearning methods. This is a problem with previous few-shot learning benchmarks. E.g. birds dataset uses very similar birds that differ only on simplistic attributes like color of beak.
We designed our benchmark to reward qualities seen in human vision: (1)Context dependence (2)Analogy Making (3)Few-shot learning with infinite vocabulary. Simply overfitting and memorization by #DeepLearning will not be enough to do well on our benchmark.
Context dependence: Same shape will have different meanings in different contexts. In left figure, highlighted shape shares a different property with other 5 positive examples, compared to right. Hence, simple pattern matching, which #DeepLearning is good at, WILL FAIL.pic.twitter.com/i8E0dOeydk
Analogy making or compositionality: Simple shapes compose together. E.g. small circles in highlighted figure are arranged together to form a "meta" shape. Humans are amazing at creating abstractions. We want to test this ability in #AIpic.twitter.com/Lz4ouU4Y9a
Infinite vocabulary: Previous benchmarks are limited to finite categories, and it is easy for #DeepLearning to memorize. To prevent this, we programmatically generate new concepts. We have subcategories: free-form, basic and abstract.pic.twitter.com/vVVi8rTCFR
Our dataset looks so simple and toy-like, yet deeply challenging for #DeepLearning due to (1)context dependence (2)abstractions (3) infinite vocabulary. We resorted to synthetic data due to severe data imbalance and scarcity with real datasets for few-shot learning.
On our benchmark, a basic neuro-symbolic method beat all neural approaches consistently and significantly. Neural approaches include latest meta/few-shot/self-supervised approaches. Shows symbol grounding is fundamental to concept learning.
Project page for our @NeurIPSConf Bongard-LOGO paper with all resources including code for dataset generation. @yukez @ZhidingYu @wn8_nie @abp4_ankit @NVIDIAAI https://research.nvidia.com/publication/2020-12_Bongard-LOGO …
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.
