• Corpus ID: 2808941

Optimal Cooperative Inference

@inproceedings{Yang2018OptimalCI,
  title={Optimal Cooperative Inference},
  author={Scott Cheng-Hsin Yang and Yue Yu and Arash Givchi and Pei Wang and Wai Keen Vong and Patrick Shafto},
  booktitle={AISTATS},
  year={2018}
}
Cooperative transmission of data fosters rapid accumulation of knowledge by efficiently combining experiences across learners. Although well studied in human learning and increasingly in machine learning, we lack formal frameworks through which we may reason about the benefits and limitations of cooperative inference. We present such a framework. We introduce novel indices for measuring the effectiveness of probabilistic and cooperative information transmission. We relate our indices to the… 

Figures from this paper

Generalizing the theory of cooperative inference
TLDR
Geometric interpretations, connections to and implications for optimal transport, and connections to importance sampling are provided, and open questions and challenges are outlined to realizing the promise of Cooperative Inference are outlined.
Bayesian Teaching of Image Categories
TLDR
The results demonstrate that Bayesian teaching can be applied to naturalistic domains, show that the best sets of examples according to the model yield better learning, and suggest avenues for improving the ability to automate teaching of image categories.
O ct 2 01 9 A mathematical theory of cooperative communication
TLDR
A mathematical theory of cooperative communication is presented that unifies three broad classes of algorithmic models as approximations of Optimal Transport, and shows that some models are provably robust to violations of common ground, even supporting online, approximate recovery from discovered violations.
Iterative Teacher-Aware Learning
TLDR
This paper gives theoretical proof that the iterative teacher-aware learning (ITAL) process leads to local and global improvements and validates the algorithms with extensive experiments on various tasks including regression, classification, and inverse reinforcement learning using synthetic and real data.
Curriculum Design for Teaching via Demonstrations: Theory and Applications
TLDR
This work provides a unified curriculum strategy for two popular learner models: Maximum Causal Entropy Inverse Reinforcement Learning (MaxEnt-IRL) and Cross-Entropy Behavioral Cloning (CrossEnt-BC) and induces a ranking over demonstrations based on a notion of difficulty scores computed w.r.t. the teacher's optimal policy and the learner’s current policy.
Supplementary Text of A mathematical theory of cooperative communication
Numerous results on SK iteration have been proved. For instance, assuming uniform marginal distributions, SK iteration of a square M converges if and only if M has at least one positive diagonal
Literal or Pedagogic Human? Analyzing Human Model Misspecification in Objective Learning
TLDR
It is argued that robots might not know whether people are being pedagogic or literal and that it is important to ask which assumption is safer to make, and it is proved that in any such game literal interpretation is more robust to misspecification.
Iterative Teaching by Label Synthesis
TLDR
This paper proposes a label synthesis teaching framework where the teacher randomly selects input teaching examples and then synthesizes suitable outputs for them and shows that this framework can avoid costly example selection while still provably achieving exponential teachability.
A mathematical theory of cooperative communication
TLDR
The results show that cooperative communication provably enables effective, robust belief transmission which is required to explain feats of human learning and improve human-machine interaction.
Sequential Cooperative Bayesian Inference
TLDR
This work seeks foundational theoretical results for cooperative inference by Bayesian agents through sequential data, and develops novel approaches analyzing consistency, rate of convergence and stability of Sequential Cooperative Bayesian Inference (SCBI).
...
1
2
...

References

SHOWING 1-10 OF 57 REFERENCES
Cooperative inference: Features, objects, and collections.
TLDR
Cooperative inference is introduced, a novel framework for cooperation in concept learning that is computationally tractable, does not require a priori agreement about biases, applies to both Boolean and first-order concepts, and begins to approximate the richness of real-world concept learning problems.
Machine Teaching for Bayesian Learners in the Exponential Family
TLDR
An optimal teaching framework aimed at learners who employ Bayesian models is proposed, expressed as an optimization problem over teaching examples that balance the future loss of the learner and the effort of the teacher.
Teaching Dimensions based on Cooperative Learning
TLDR
The model of subset teaching sets is introduced, based on the idea that both teacher and learner can exploit the assumption that the partner is cooperative, and it is shown how this can reduce the sample size drastically without using coding tricks.
Showing versus doing: Teaching by demonstration
TLDR
A novel Bayesian model for teaching by demonstration is developed and it is shown that human participants systematically modify their teaching behavior consistent with the predictions of the model.
Multi-Agent Actor-Critic for Mixed Cooperative-Competitive Environments
TLDR
An adaptation of actor-critic methods that considers action policies of other agents and is able to successfully learn policies that require complex multi-agent coordination is presented.
Learning From Others
TLDR
A Bayesian analysis of learning from knowledgeable others is provided, which formalizes how learners may use a person’s actions and goals to make inferences about the actor's knowledge about the world.
Cooperative Inverse Reinforcement Learning
TLDR
It is shown that computing optimal joint policies in CIRL games can be reduced to solving a POMDP, it is proved that optimality in isolation is suboptimal in C IRL, and an approximate CirL algorithm is derived.
Optimal Teaching for Limited-Capacity Human Learners
TLDR
This work applies a machine teaching procedure to a cognitive model that is either limited capacity ( as humans are) or unlimited capacity (as most machine learning systems are), and finds that the machine teacher recommends idealized training sets and human learners perform best when training recommendations are based on a limited-capacity model.
Cooperating with machines
TLDR
An algorithm is developed that can cooperate effectively with humans when cooperation is beneficial but nontrivial, something humans are remarkably good at, and indicates that general human–machine cooperation is achievable using a non-trivial but ultimately simple set of algorithmic mechanisms.
...
1
2
3
4
5
...