SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2008.11297 · NeurIPS · 2020

Transductive Information Maximization For Few-Shot Learning

Jose Dolz, Malik Boudiaf, Éts Montreal, Ziko Masud, Jérôme Rony, Pablo Piantanida, Ismail Ben, Ayed Montreal

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 10 functions out of this paper's own repositories and ran 9 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
mboudiaf/TIM canonical 6 of 7
sicara/easy-few-shot-learning — 3 of 3
FunctionStatusWhere it lives
FewShotClassifier Ran sicara/easy-few-shot-learning/easyfsl/methods/tim.py
code served (permissive licence) · get_code("637e871cf76bfc10")
TIM Ran sicara/easy-few-shot-learning/easyfsl/methods/tim.py
code served (permissive licence) · get_code("c46ba9a1cacbe109")
compute_prototypes Ran sicara/easy-few-shot-learning/easyfsl/methods/tim.py
code served (permissive licence) · get_code("31e41d7e676cb492")
conv_block Ran mboudiaf/TIM/src/models/Conv4.py
code served (permissive licence) · get_code("02652b1ac1a54dda")
get_features Ran mboudiaf/TIM/src/utils.py
code served (permissive licence) · get_code("12cbd4d101797d30")
get_metric Ran mboudiaf/TIM/src/models/ProtoNet.py
code served (permissive licence) · get_code("235da783eeb0573f")
get_one_hot Ran mboudiaf/TIM/src/utils.py
code served (permissive licence) · get_code("a03c85ae21c3b667")
with_augment Ran mboudiaf/TIM/src/datasets/transform.py
code served (permissive licence) · get_code("97533e04cdf4da7e")
without_augment Ran mboudiaf/TIM/src/datasets/transform.py
code served (permissive licence) · get_code("e1f7a44f9ac823d7")
get_logs_path Not yet run mboudiaf/TIM/src/utils.py
code served (permissive licence) · get_code("e232226ad7c9556f")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We introduce Transductive Infomation Maximization (TIM) for few-shot learning. Our method maximizes the mutual information between the query features and their label predictions for a given few-shot task, in conjunction with a supervision loss based on the support set. Furthermore, we propose a new alternating-direction solver for our mutual-information loss, which substantially speeds up transductiveinference convergence over gradient-based optimization, while yielding similar accuracy. TIM inference is modular: it can be used on top of any base-training feature extractor. Following standard transductive few-shot settings, our comprehensive experiments 2 demonstrate that TIM outperforms state-of-the-art methods significantly across various datasets and networks, while used on top of a fixed feature extractor trained with simple cross-entropy on the base classes, without resorting to complex meta-learning schemes. It consistently brings between 2% and 5% improvement in accuracy over the best performing method, not only on all the well-established few-shot benchmarks but also on more challenging scenarios, with domain shifts and larger numbers of classes.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2008.11297")
get_code_for_paper("2008.11297")
have("2008.11297")

Connect an agent — have() is free.