Do Not Marginalize Mechanisms, Rather Consolidate!
Authors: Moritz Willig, Matej Zečević, Devendra Dhami, Kristian Kersting
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate consolidation on two examples. First, obtaining a causal model of reduced size and, secondly, revealing the underlying policy of a causal decision making process. ... Figure 4 shows the resulting plot of the consolidated model under interventions do(S12 = 1) and do(S24 = 1). We successfully demonstrated the power of consolidation models for dynamical system while preserving the ability to intervene. In our second example we apply consolidation to a more complex causal graph relating the game state of a simple platformer environment to the actions of an agent. |
| Researcher Affiliation | Academia | Moritz Willig Technical University of Darmstadt moritz.willig@cs.tu-darmstadt.de Matej Zeˇcevi c Technical University of Darmstadt matej.zecevic@tu-darmstadt.de Devendra Singh Dhami Eindhoven University of Technology Hessian Center for AI (hessian.AI) d.s.dhami@tue.nl Kristian Kersting Technical University Darmstadt Hessian Center for AI (hessian.AI) German Research Center for AI (DFKI) kersting@cs.tu-darmstadt.de |
| Pseudocode | Yes | Figure 3: CONSOLIDATE Algorithm. The above pseudo-code summarizes the consolidation algorithm as described in this paper by utilizing causal compositional variables and partitioned SCM to obtain simplified SCM. |
| Open Source Code | No | The paper does not contain an explicit statement about releasing source code for the described methodology or a direct link to a code repository. |
| Open Datasets | No | The paper describes examples like 'time series data' and 'game agent policy' with their SCMs, but does not provide concrete access information (links, DOIs, specific citations with authors/year) for publicly available datasets used for training or evaluation. |
| Dataset Splits | No | The paper does not provide specific dataset split information (exact percentages, sample counts, citations to predefined splits, or detailed splitting methodology) needed to reproduce data partitioning. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, memory amounts, or detailed computer specifications) used for running its experiments. |
| Software Dependencies | No | The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers like Python 3.8, CPLEX 12.4) needed to replicate the experiment. |
| Experiment Setup | Yes | We observe the tool to loose some percentage of its sharpness per day depending on its utilization Ut. The intervention do(St = 1) resets the sharpness to a certain sharpness. ... U = {Ut = N(0.5, 0.05^2)} V = {Lt, St, At} I = P({do(St = 1), do(Lt = 1.0), do(St = 1, Lt = 1)}) fl(l, u) := (1.0 - 0.002u)l fst(st - 1, u) := (1.0 - 0.3u)st - 1 fa(s) := 0.8s^2 |