-
Diffusion Models as Artists: Are we Closing the Gap between Humans and Machines?
Authors:
Victor Boutin,
Thomas Fel,
Lakshya Singhal,
Rishav Mukherji,
Akash Nagaraj,
Julien Colin,
Thomas Serre
Abstract:
An important milestone for AI is the development of algorithms that can produce drawings that are indistinguishable from those of humans. Here, we adapt the 'diversity vs. recognizability' scoring framework from Boutin et al, 2022 and find that one-shot diffusion models have indeed started to close the gap between humans and machines. However, using a finer-grained measure of the originality of in…
▽ More
An important milestone for AI is the development of algorithms that can produce drawings that are indistinguishable from those of humans. Here, we adapt the 'diversity vs. recognizability' scoring framework from Boutin et al, 2022 and find that one-shot diffusion models have indeed started to close the gap between humans and machines. However, using a finer-grained measure of the originality of individual samples, we show that strengthening the guidance of diffusion models helps improve the humanness of their drawings, but they still fall short of approximating the originality and recognizability of human drawings. Comparing human category diagnostic features, collected through an online psychophysics experiment, against those derived from diffusion models reveals that humans rely on fewer and more localized features. Overall, our study suggests that diffusion models have significantly helped improve the quality of machine-generated drawings; however, a gap between humans and machines remains -- in part explainable by discrepancies in visual strategies.
△ Less
Submitted 31 May, 2023; v1 submitted 27 January, 2023;
originally announced January 2023.
-
Diversity vs. Recognizability: Human-like generalization in one-shot generative models
Authors:
Victor Boutin,
Lakshya Singhal,
Xavier Thomas,
Thomas Serre
Abstract:
Robust generalization to new concepts has long remained a distinctive feature of human intelligence. However, recent progress in deep generative models has now led to neural architectures capable of synthesizing novel instances of unknown visual concepts from a single training example. Yet, a more precise comparison between these models and humans is not possible because existing performance metri…
▽ More
Robust generalization to new concepts has long remained a distinctive feature of human intelligence. However, recent progress in deep generative models has now led to neural architectures capable of synthesizing novel instances of unknown visual concepts from a single training example. Yet, a more precise comparison between these models and humans is not possible because existing performance metrics for generative models (i.e., FID, IS, likelihood) are not appropriate for the one-shot generation scenario. Here, we propose a new framework to evaluate one-shot generative models along two axes: sample recognizability vs. diversity (i.e., intra-class variability). Using this framework, we perform a systematic evaluation of representative one-shot generative models on the Omniglot handwritten dataset. We first show that GAN-like and VAE-like models fall on opposite ends of the diversity-recognizability space. Extensive analyses of the effect of key model parameters further revealed that spatial attention and context integration have a linear contribution to the diversity-recognizability trade-off. In contrast, disentanglement transports the model along a parabolic curve that could be used to maximize recognizability. Using the diversity-recognizability framework, we were able to identify models and parameters that closely approximate human data.
△ Less
Submitted 7 October, 2022; v1 submitted 20 May, 2022;
originally announced May 2022.
-
On the coefficient-choosing game
Authors:
Divyum Sharma,
L. Singhal
Abstract:
Nora and Wanda are two players who choose coefficients of a degree $d$ polynomial from some fixed unital commutative ring $R$. Wanda is declared the winner if the polynomial has a root in the ring of fractions of $R$ and Nora is declared the winner otherwise. We extend the theory of these games given by Gasarch, Washington and Zbarsky to all finite cyclic rings and determine the possible outcomes.…
▽ More
Nora and Wanda are two players who choose coefficients of a degree $d$ polynomial from some fixed unital commutative ring $R$. Wanda is declared the winner if the polynomial has a root in the ring of fractions of $R$ and Nora is declared the winner otherwise. We extend the theory of these games given by Gasarch, Washington and Zbarsky to all finite cyclic rings and determine the possible outcomes. A family of examples is also constructed using discrete valuation rings for a variant of the game proposed by these authors. Our techniques there lead us to an adversarial approach to constructing rational polynomials of any prescribed degree (equal to $3$ or greater than $8$) with no roots in the maximal abelian extension of $\mathbb{Q}$.
△ Less
Submitted 30 November, 2020; v1 submitted 30 June, 2020;
originally announced July 2020.
-
Inhomogeneous Diophantine approximation over fields of formal power series
Authors:
Yann Bugeaud,
L. Singhal,
Zhenliang Zhang
Abstract:
We prove a sharp analogue of Minkowski's inhomogeneous approximation theorem over fields of power series $\mathbb{F}_q((T^{-1}))$. Furthermore, we study the approximation to a given point $\underline{y}$ in $\mathbb{F}_q((T^{-1}))^2$ by the $SL_2(\mathbb{F}_q[T])$-orbit of a given point $\underline{x}$ in $\mathbb{F}_q((T^{-1}))^2$.
We prove a sharp analogue of Minkowski's inhomogeneous approximation theorem over fields of power series $\mathbb{F}_q((T^{-1}))$. Furthermore, we study the approximation to a given point $\underline{y}$ in $\mathbb{F}_q((T^{-1}))^2$ by the $SL_2(\mathbb{F}_q[T])$-orbit of a given point $\underline{x}$ in $\mathbb{F}_q((T^{-1}))^2$.
△ Less
Submitted 4 September, 2019;
originally announced September 2019.
-
Cylinder absolute games on solenoids
Authors:
L. Singhal
Abstract:
Let $A$ be any affine surjective endomorphism of a solenoid $Σ_{\mathcal{P}}$ over the circle $S^1$ which is not an infinite-order translation of $Σ_{\mathcal{P}}$. We prove the existence of a cylinder absolute winning (CAW) subset $F \subset Σ_{\mathcal{P}}$ with the property that for any $x \in F$, the orbit closure $\overline{\{ A^{\ell} x \mid \ell \in \mathbb{N} \}}$ does not contain any peri…
▽ More
Let $A$ be any affine surjective endomorphism of a solenoid $Σ_{\mathcal{P}}$ over the circle $S^1$ which is not an infinite-order translation of $Σ_{\mathcal{P}}$. We prove the existence of a cylinder absolute winning (CAW) subset $F \subset Σ_{\mathcal{P}}$ with the property that for any $x \in F$, the orbit closure $\overline{\{ A^{\ell} x \mid \ell \in \mathbb{N} \}}$ does not contain any periodic orbits. The class of infinite solenoids considered in this paper provides, to our knowledge, some of the first examples of non-Federer spaces where absolute games can be played and won. Dimension maximality and incompressibility of CAW sets is also discussed for a number of possibilities in addition to their winning nature for the games known from before.
△ Less
Submitted 30 June, 2019; v1 submitted 24 May, 2018;
originally announced May 2018.
-
Diophantine exponents for standard linear actions of $\mathrm{SL}_2$ over discrete rings in $\mathbb{C}$
Authors:
L. Singhal
Abstract:
We give upper and lower bounds for various Diophantine exponents associated with the standard linear actions of $\mathrm{SL}_2 ( \mathcal{O}_K )$ on the punctured complex plane $\mathbb{C}^2 \setminus \{ \mathbf{0} \}$, where $K$ is a number field whose ring of integers $\mathcal{O}_K$ is discrete and within a unit distance of any complex number. The results are similar to those of Laurent and Nog…
▽ More
We give upper and lower bounds for various Diophantine exponents associated with the standard linear actions of $\mathrm{SL}_2 ( \mathcal{O}_K )$ on the punctured complex plane $\mathbb{C}^2 \setminus \{ \mathbf{0} \}$, where $K$ is a number field whose ring of integers $\mathcal{O}_K$ is discrete and within a unit distance of any complex number. The results are similar to those of Laurent and Nogueira for $\mathrm{SL}_2 ( \mathbb{Z} )$ action on $\mathbb{R}^2 \setminus \{ \mathbf{0} \}$ albeit for us, uniformly nice bounds are obtained only outside of a set of null measure.
△ Less
Submitted 4 September, 2015;
originally announced September 2015.
