Tom McCoy

@rtommccoy

Assistant professor @YaleLinguistics. Studying computational linguistics, cognitive science, and AI. He/him.

following
followers

The case for neural networks as approximate symbolic systems

A researcher describes the work as a review-and-opinion paper on symbolic representations, algorithms and learning in AI networks.

8 Sources, 6d ago, updated 6d ago

Top: The title, authors, and journal of the paper. Title: "Reconciling symbols and neural networks: Evidence for a quasi-symbolic view." Authors: "Tom McCoy, Abi Tenenbaum, Zhang Enyan, Zhenghao Herbert Zhou." Journal: "Current Opinion in Behavioral Sciences"

Bottom: A figure displaying how the vector representations learned by neural networks often show emergent symbolic structure. For instance, they can often be decomposed into meaningful constituents (illustrated in the figure by adding constituents to make a vector) or classified in vector space along symbolic dimensions (shown by an image of linear classifiers). However, these vectors are only quasi-symbolic: they deviate from pure symbolic structure in important ways - e.g., they can display graded similarity (“movie” is closer to “novel” than to “meal”), and they may capture graded context effects such as the word “good” affecting different nouns differently (the image shows such graded similarity among points in space

What remains unexplained about AI generalization?

A user asks what remains mysterious about generalization; a reply points to defining “domain” and distinguishing generalization from memorization.

لY

9 Sources, 5d ago, updated 5d ago

Illustration of meta-learning. A network starts with initial parameters theta and can then be rapidly adapted to learn any formal pattern of the form (A B)^n; e.g., (1 6)^n means any string made of n copies of 1 6, such as “1 6”, “1 6 1 6”, “1 6 1 6 1 6”, etc. The parameters that result from this rapid adaptation show additive structure; e.g., the parameters for (1 5)^n minus the parameters for (2 6)^n plus the parameters for (2 5)^n produces parameters that yield (1 6)^n

Study finds symbolic structure inside large language model representations

A researcher likens vectors inside large language models to pointillist paintings while sharing Yale News coverage of their work.

1 Source, 12d ago, first seen 12d ago

LLM representations are claimed to have implicit symbolic structure

A researcher says the finding follows an eight-year project into how LLMs handle language, code and math.

1 Source, 12d ago, first seen 12d ago

Overview of the paper.
Title: The Emergent Symbolic Structure of Artificial Neural Networks
Authors: Tom McCoy, Paul Soulos, Tal Linzen, Paul Smolensky
Left: Neural networks encode information in vectors (there is then an image of a vector), yet they excel at tasks long thought to require symbolic structure (there is then an image of a symbolic representation, specifically a syntax tree). How do LLMs do it?
Right: We find that LLM representations can be closely approximated with symbolic structures. This approximation lets us edit the structure of an LLM’s output by editing the structure of its internal representations, as shown. There is then an image of two edits to LLMs. In the first one, the original input is 3 + 6 * 8, with an answer of 51. But if we swap the positions of the 3 and the 6, the output becomes 30. In the second one, the original input is a Python command repeating the list [Z, U] three times, producing [Z, U, Z, U, Z, U]. But if we edit the input in a way that adds

LLM verb learning reportedly depends more on syntax than co-occurrence cues

A researcher says their new paper on Open Mind investigates the roles of syntactic and co-occurrence cues in how large language models learn verb meanings.

M筱

5 Sources, 21d ago, updated 21d ago

4 Sources, 46d ago, first seen 46d ago

Tom McCoy Discusses TPR-Augmented Transformers

Tom McCoy references a paper on augmenting Transformers with TPR structure.

3 Sources, 49d ago, first seen 49d ago

Tom McCoy Posts Paper on Emergent Symbolic Structure in Neural Networks

Yale assistant professor posts arXiv paper and thread testing implicit tensor product representations in neural networks.

34 Sources, 49d ago, first seen 49d ago

Tom McCoy

The case for neural networks as approximate symbolic systems

What remains unexplained about AI generalization?

Study finds symbolic structure inside large language model representations

LLM representations are claimed to have implicit symbolic structure

LLM verb learning reportedly depends more on syntax than co-occurrence cues

NYU’s computational cognitive science and AI community

Tom McCoy on Compositional Filler-Role Representations

New Paper Unifies Interpretability Methods via Tensor Product Representations

Tom McCoy Discusses TPR-Augmented Transformers

Tom McCoy Posts Paper on Emergent Symbolic Structure in Neural Networks

Tom McCoy

The case for neural networks as approximate symbolic systems

What remains unexplained about AI generalization?

Study finds symbolic structure inside large language model representations

LLM representations are claimed to have implicit symbolic structure

LLM verb learning reportedly depends more on syntax than co-occurrence cues

NYU’s computational cognitive science and AI community

Tom McCoy on Compositional Filler-Role Representations

New Paper Unifies Interpretability Methods via Tensor Product Representations

Tom McCoy Discusses TPR-Augmented Transformers

Tom McCoy Posts Paper on Emergent Symbolic Structure in Neural Networks

Featured in

The case for neural networks as approximate symbolic systems

What remains unexplained about AI generalization?

Study finds symbolic structure inside large language model representations

LLM representations are claimed to have implicit symbolic structure

LLM verb learning reportedly depends more on syntax than co-occurrence cues

NYU’s computational cognitive science and AI community

Tom McCoy on Compositional Filler-Role Representations

New Paper Unifies Interpretability Methods via Tensor Product Representations

Tom McCoy Discusses TPR-Augmented Transformers

Tom McCoy Posts Paper on Emergent Symbolic Structure in Neural Networks

Featured in

The case for neural networks as approximate symbolic systems

What remains unexplained about AI generalization?

Study finds symbolic structure inside large language model representations

LLM representations are claimed to have implicit symbolic structure

LLM verb learning reportedly depends more on syntax than co-occurrence cues

NYU’s computational cognitive science and AI community

Tom McCoy on Compositional Filler-Role Representations

New Paper Unifies Interpretability Methods via Tensor Product Representations

Tom McCoy Discusses TPR-Augmented Transformers

Tom McCoy Posts Paper on Emergent Symbolic Structure in Neural Networks