HOUDINI: Lifelong Learning as Program Synthesis

By Lazar Valkov, Dipak Chaudhari, Akash Srivastava, Charles Sutton, Swarat Chaudhuri. 2018

86 citations recorded by the source corpus.

We present a neurosymbolic framework for the lifelong learning of algorithmic tasks that mix perception and procedural reasoning. Reusing high-level concepts across domains and learning complex procedures are key challenges in lifelong learning. We show that a program synthesis approach that combines gradient descent with combinatorial search over programs can be a more effective response to these challenges than purely neural methods. Our framework, called HOUDINI, represents neural networks as strongly typed, differentiable functional programs that use symbolic higher-order combinators to compose a library of neural functions. Our learning algorithm consists of: (1) a symbolic program synthesizer that performs a type-directed search over parameterized programs, and decides on the library functions to reuse, and the architectures to combine them, while learning a sequence of tasks; and (2) a neural module that trains these programs using stochastic gradient descent. We evaluate HOUDINI on three benchmarks that combine perception with the algorithmic tasks of counting, summing, and shortest-path computation. Our experiments show that HOUDINI transfers high-level concepts more effectively than traditional transfer learning and progressive neural networks, and that the typed representation of networks signiﬁcantly accelerates the search.

Subfields: Artificial Intelligence, Electrical and Electronic Engineering, Hardware and Architecture

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Electrical and Electronic Engineering

In Hardware and Architecture

Citations in the corpus

Compacting, Picking and Growing for Unforgetting Continual Learning (2019, background): “Besides, works on task-free [2] or as a program synthesis [43] have also been studied recently.”
Automatically Composing Representation Transformations as a Means for Generalization (2018, background): “ approaches are respectively analogous to our hardcoded-functions and hardcoded-controller ablations in Fig. 7. The closest works to ours from a program induction perspective are (Gaunt et al., 2016; Valkov et al., 2018), both neurosymbolic approaches for learning differentiable programs integrated ”
Lifelong Learning of Compositional Structures (2020, background): “Most work in this line assumes that each component can be fully learned by training on a single task, and then can be reused for other tasks [24, 7, 29].”
How to Reuse and Compose Knowledge for a Lifetime of Tasks: A Survey on Continual Learning and Functional Composition (2022, background): “Some such approaches introduce heuristics for searching over the space of possible module configurations upon encountering a new task to improve efficiency, for example using programming languages techniques (Valkov et al., 2018) or data-driven heuristics (Veniat et al.”
Task-Driven Modular Networks for Zero-Shot Compositional Learning (2019, methodology): “These networks have been used in the past to speed up computation at test time [1] and to improve generalization for multi-task learning [19, 26], reinforcement learning [7], continual learning [30], visual question answering [2, 25],”
Learning Interpretable Models with Causal Guarantees (2019, unknown): “…Rudin, 2016), generalized additive models (Lou et al., 2012; Caruana et al., 2015), rule lists (Wang and Rudin, 2015; Yang et al., 2017; Angelino et al., 2017), decision sets (Lakkaraju et al., 2016), and programs (Ellis et al., 2015; Verma et al., 2018; Valkov et al., 2018; Ellis et al., 2018).”
A Probabilistic Framework for Modular Continual Learning (2023, methodology): “We identify the following CL desiderata [41, 44]: 1.”
Neurosymbolic Program Synthesis (unknown): “For example, the H OUDINI system [9] uses a search-and-train approach.”
Neurosymbolic Programming (2021, methodology): “The Houdini approach (Valkov et al., 2018) uses this idea to perform structured computations over natural inputs, for example, to count the number of cars in a photograph.”
DreamCoder: bootstrapping inductive program synthesis with wake-sleep library learning (2021, background): “Further combining library learning with hybrid discrete/continuous or neurosymbolic program representations [56] could be valuable, especially for scaling to real-world domains with messier structure than those considered here.”