Figure 4 - uploaded by Tomasz Odrzygóźdź
Content may be subject to copyright.
Download
View publication
Histogram of ∆. Note that 17% of subgoals increases the distance. Additional, 5% leads to unsolvable "dead states" present in Sokoban.

Histogram of ∆. Note that 17% of subgoals increases the distance. Additional, 5% leads to unsolvable "dead states" present in Sokoban.

Source publication
Figure 1: The performance of Subgoal Search. (top, left) comparison on...
Figure 2: Comparison of success rates w.r.t k.
Figure 3: Out-of-distribution generalization to longer proofs. We...
Figure 4: Histogram of ∆. Note that 17% of subgoals increases the...
+2 Table 4 .
Subgoal Search For Complex Reasoning Tasks
Preprint
Full-text available
  • Aug 2021
Humans excel in solving complex reasoning tasks through a mental process of moving from one idea to a related one. Inspired by this, we propose Subgoal Search (kSubS) method. Its key component is a learned subgoal generator that produces a diversity of subgoals that are both achievable and closer to the solution. Using subgoals reduces the search s...
Cite
Download full-text

Contexts in source publication

Context 1
... study this on 10 × 10 boards of Sokoban, which are small enough to calculate the true distance dist to the solution using the Dijkstra algorithm. In Figure 4, we study ∆ := dist s1 − dist s2 , where s 1 is a sampled state and s 2 is a subgoal generated from s 1 . Ideally, the histogram should concentrate on k = 4 used in training. ...
Context 2
... study this on 10 × 10 boards of Sokoban, which are small enough to calculate the true distance dist to the solution using the Dijkstra algorithm. In Figure 4, we study ∆ := dist s1 − dist s2 , where s 1 is a sampled state and s 2 is a subgoal generated from s 1 . Ideally, the histogram should concentrate on k = 4 used in training. ...

Similar publications

Fig. 3. The overall score θ with γ = 0.5
Fig. 4. An instance of HRG model. In (a), the red and black edges are...
Fig. 6. Model performance of the feature-level protection.
Fig. 7. Model performance of the edge-level protection.
Heterogeneous Randomized Response for Differential Privacy in Graph Neural Networks
Preprint
Full-text available
  • Nov 2022
Graph neural networks (GNNs) are susceptible to privacy inference attacks (PIAs), given their ability to learn joint representation from features and edges among nodes in graph data. To prevent privacy leakages in GNNs, we propose a novel heterogeneous randomized response (HeteroRR) mechanism to protect nodes' features and edges against PIAs under...
View
Fig. 2. An instance of HRG model (b) represents a given graph G (a)....
Fig. 3. The overall score θ with γ = 0.5
Fig. 4. An instance of HRG model. In (a), the red and black edges are...
Fig. 6. Model performance of the feature-level protection.
Fig. 7. Model performance of the edge-level protection.
Heterogeneous Randomized Response for Differential Privacy in Graph Neural Networks
Conference Paper
Full-text available
  • Nov 2022
Graph neural networks (GNNs) are susceptible to privacy inference attacks (PIAS) given their ability to learn joint representation from features and edges among nodes in graph data. To prevent privacy leakages in GNNs, we propose a novel heterogeneous randomized response (HETERORR) mechanism to protect nodes' features and edges against PIAS under d...
View