Showing 1–10 of 10 results for author: Assassi, V

Rewarding Chatbots for Real-World Engagement with Millions of Users

Authors: Robert Irvine, Douglas Boubert, Vyas Raina, Adian Liusie, Ziyi Zhu, Vineet Mudupalli, Aliaksei Korshuk, Zongyi Liu, Fritz Cremer, Valentin Assassi, Christie-Carol Beauchamp, Xiaoding Lu, Thomas Rialan, William Beauchamp

Abstract: The emergence of pretrained large language models has led to the deployment of a range of social chatbots for chitchat. Although these chatbots demonstrate language ability and fluency, they are not guaranteed to be engaging and can struggle to retain users. This work investigates the development of social chatbots that prioritize user engagement to enhance retention, specifically examining the us… ▽ More The emergence of pretrained large language models has led to the deployment of a range of social chatbots for chitchat. Although these chatbots demonstrate language ability and fluency, they are not guaranteed to be engaging and can struggle to retain users. This work investigates the development of social chatbots that prioritize user engagement to enhance retention, specifically examining the use of human feedback to efficiently develop highly engaging chatbots. The proposed approach uses automatic pseudo-labels collected from user interactions to train a reward model that can be used to reject low-scoring sample responses generated by the chatbot model at inference time. Intuitive evaluation metrics, such as mean conversation length (MCL), are introduced as proxies to measure the level of engagement of deployed chatbots. A/B testing on groups of 10,000 new daily chatbot users on the Chai Research platform shows that this approach increases the MCL by up to 70%, which translates to a more than 30% increase in user retention for a GPT-J 6B model. Future work aims to use the reward model to realise a data fly-wheel, where the latest user conversations can be used to alternately fine-tune the language model and the reward model. △ Less

Submitted 30 March, 2023; v1 submitted 10 March, 2023; originally announced March 2023.

Modeling Biased Tracers at the Field Level

Authors: Marcel Schmittfull, Marko Simonović, Valentin Assassi, Matias Zaldarriaga

Abstract: In this paper we test the perturbative halo bias model at the field level. The advantage of this approach is that any analysis can be done without sample variance if the same initial conditions are used in simulations and perturbation theory calculations. We write the bias expansion in terms of modified bias operators in Eulerian space, designed such that the large bulk flows are automatically res… ▽ More In this paper we test the perturbative halo bias model at the field level. The advantage of this approach is that any analysis can be done without sample variance if the same initial conditions are used in simulations and perturbation theory calculations. We write the bias expansion in terms of modified bias operators in Eulerian space, designed such that the large bulk flows are automatically resummed and not treated perturbatively. Using these operators, the bias model accurately matches the Eulerian density of halos in N-body simulations. The mean-square model error is close to the Poisson shot noise for a wide range of halo masses and it is rather scale-independent, with scale-dependent corrections becoming relevant at the nonlinear scale. In contrast, for linear bias the mean-square model error can be higher than the Poisson prediction by factors of up to a few on large scales, and it becomes scale dependent already in the linear regime. We show that by weighting simulated halos by their mass, the mean-square error of the model can be further reduced by up to an order of magnitude, or by a factor of two when including $60\%$ mass scatter. We also test the Standard Eulerian bias model using the nonlinear matter field measured from simulations and show that it leads to a larger and more scale-dependent model error than the bias expansion based on perturbation theory. These results may be of particular relevance for cosmological inference methods that use a likelihood of the biased tracer at the field level, or for initial condition and BAO reconstruction that requires a precise estimate of the large-scale potential from the biased tracer density. △ Less

Submitted 4 July, 2019; v1 submitted 26 November, 2018; originally announced November 2018.

Comments: 61 pages, 27 figures. Minor edits and added references to match published version

Journal ref: Phys. Rev. D 100, 043514 (2019)

Efficient Evaluation of Cosmological Angular Statistics

Authors: Valentin Assassi, Marko Simonović, Matias Zaldarriaga

Abstract: Angular statistics of cosmological observables are hard to compute. The main difficulty is due to the presence of highly-oscillatory Bessel functions which need to be integrated over. In this paper, we provide a simple and fast method to compute the angular power spectrum and bispectrum of any observable. The method is based on using an FFTlog algorithm to decompose the momentum-space statistics o… ▽ More Angular statistics of cosmological observables are hard to compute. The main difficulty is due to the presence of highly-oscillatory Bessel functions which need to be integrated over. In this paper, we provide a simple and fast method to compute the angular power spectrum and bispectrum of any observable. The method is based on using an FFTlog algorithm to decompose the momentum-space statistics onto a basis of power-law functions. For each power law, the integrals over Bessel functions have a simple analytical solution. This allows us to efficiently evaluate these integrals, independently of the value of the multipole $\ell$. In particular, this method significantly speeds up the evaluation of the angular bispectrum compared to existing methods. To illustrate our algorithm, we compute the galaxy, lensing and CMB temperature angular power spectrum and bispectrum. △ Less

Submitted 8 December, 2017; v1 submitted 14 May, 2017; originally announced May 2017.

Comments: 31 pages, 10 figures, 2 tables. Added section on redshift space distortions. Added section on parameters and performances in appendix. Several minor modifications to match the published version in JCAP. (Title in published version changed)

Galaxy Bias and Primordial Non-Gaussianity

Authors: Valentin Assassi, Daniel Baumann, Fabian Schmidt

Abstract: We present a systematic study of galaxy biasing in the presence of primordial non-Gaussianity. For a large class of non-Gaussian initial conditions, we define a general bias expansion and prove that it is closed under renormalization, thereby showing that the basis of operators in the expansion is complete. We then study the effects of primordial non-Gaussianity on the statistics of galaxies. We s… ▽ More We present a systematic study of galaxy biasing in the presence of primordial non-Gaussianity. For a large class of non-Gaussian initial conditions, we define a general bias expansion and prove that it is closed under renormalization, thereby showing that the basis of operators in the expansion is complete. We then study the effects of primordial non-Gaussianity on the statistics of galaxies. We show that the equivalence principle enforces a relation between the scale-dependent bias in the galaxy power spectrum and that in the dipolar part of the bispectrum. This provides a powerful consistency check to confirm the primordial origin of any observed scale-dependent bias. Finally, we also discuss the imprints of anisotropic non-Gaussianity as motivated by recent studies of higher-spin fields during inflation. △ Less

Submitted 25 November, 2015; v1 submitted 13 October, 2015; originally announced October 2015.

Comments: 37 pages, 2 figures; v2: minor clarifications in Section 3

Effective Theory of Large-Scale Structure with Primordial Non-Gaussianity

Authors: Valentin Assassi, Daniel Baumann, Enrico Pajer, Yvette Welling, Drian van der Woude

Abstract: We develop the effective theory of large-scale structure for non-Gaussian initial conditions. The effective stress tensor in the dark matter equations of motion contains new operators, which originate from the squeezed limit of the primordial bispectrum. Parameterizing the squeezed limit by a scaling and an angular dependence, captures large classes of primordial non-Gaussianity. Within this param… ▽ More We develop the effective theory of large-scale structure for non-Gaussian initial conditions. The effective stress tensor in the dark matter equations of motion contains new operators, which originate from the squeezed limit of the primordial bispectrum. Parameterizing the squeezed limit by a scaling and an angular dependence, captures large classes of primordial non-Gaussianity. Within this parameterization, we classify the possible contributions to the effective theory. We show explicitly how all terms consistent with the symmetries arise from coarse graining the dark matter equations of motion and its initial conditions. We also demonstrate that the system is closed under renormalization and that the basis of correction terms is therefore complete. The relevant corrections to the matter power spectrum and bispectrum are computed numerically and their relative importance is discussed. △ Less

Submitted 25 May, 2015; originally announced May 2015.

Comments: 59 pages, 33 figures

Testing Inflation with Large Scale Structure: Connecting Hopes with Reality

Authors: Marcelo Alvarez, Tobias Baldauf, J. Richard Bond, Neal Dalal, Roland de Putter, Olivier Doré, Daniel Green, Chris Hirata, Zhiqi Huang, Dragan Huterer, Donghui Jeong, Matthew C. Johnson, Elisabeth Krause, Marilena Loverde, Joel Meyers, P. Daniel Meerburg, Leonardo Senatore, Sarah Shandera, Eva Silverstein, Anže Slosar, Kendrick Smith, Matias Zaldarriaga, Valentin Assassi, Jonathan Braden, Amir Hajian , et al. (3 additional authors not shown)

Abstract: The statistics of primordial curvature fluctuations are our window into the period of inflation, where these fluctuations were generated. To date, the cosmic microwave background has been the dominant source of information about these perturbations. Large scale structure is however from where drastic improvements should originate. In this paper, we explain the theoretical motivations for pursuing… ▽ More The statistics of primordial curvature fluctuations are our window into the period of inflation, where these fluctuations were generated. To date, the cosmic microwave background has been the dominant source of information about these perturbations. Large scale structure is however from where drastic improvements should originate. In this paper, we explain the theoretical motivations for pursuing such measurements and the challenges that lie ahead. In particular, we discuss and identify theoretical targets regarding the measurement of primordial non-Gaussianity. We argue that when quantified in terms of the local (equilateral) template amplitude $f_{\rm NL}^{\rm loc}$ ($f_{\rm NL}^{\rm eq}$), natural target levels of sensitivity are $Δf_{\rm NL}^{\rm loc, eq.} \simeq 1$. We highlight that such levels are within reach of future surveys by measuring 2-, 3- and 4-point statistics of the galaxy spatial distribution. This paper summarizes a workshop held at CITA (University of Toronto) on October 23-24, 2014. △ Less

Submitted 15 December, 2014; originally announced December 2014.

Comments: 27 pages + references

Renormalized Halo Bias

Authors: Valentin Assassi, Daniel Baumann, Daniel Green, Matias Zaldarriaga

Abstract: This paper provides a systematic study of renormalization in models of halo biasing. Building on work of McDonald, we show that Eulerian biasing is only consistent with renormalization if non-local terms and higher-derivative contributions are included in the biasing model. We explicitly determine the complete list of required bias parameters for Gaussian initial conditions, up to quartic order in… ▽ More This paper provides a systematic study of renormalization in models of halo biasing. Building on work of McDonald, we show that Eulerian biasing is only consistent with renormalization if non-local terms and higher-derivative contributions are included in the biasing model. We explicitly determine the complete list of required bias parameters for Gaussian initial conditions, up to quartic order in the dark matter density contrast and at leading order in derivatives. At quadratic order, this means including the gravitational tidal tensor, while at cubic order the velocity potential appears as an independent degree of freedom. Our study naturally leads to an effective theory of biasing in which the halo density is written as a double expansion in fluctuations and spatial derivatives. We show that the bias expansion can be organized in terms of Galileon operators which aren't renormalized at leading order in derivatives. Finally, we discuss how the renormalized bias parameters impact the statistics of halos. △ Less

Submitted 24 February, 2014; originally announced February 2014.

Comments: 44 pages, 6 figures

Planck-Suppressed Operators

Authors: Valentin Assassi, Daniel Baumann, Daniel Green, Liam McAllister

Abstract: We show that the recent Planck limits on primordial non-Gaussianity impose strong constraints on light hidden sector fields coupled to the inflaton via operators suppressed by a high mass scale Λ. We study a simple effective field theory in which a hidden sector field is coupled to a shift-symmetric inflaton via arbitrary operators up to dimension five. Self-interactions in the hidden sector lead… ▽ More We show that the recent Planck limits on primordial non-Gaussianity impose strong constraints on light hidden sector fields coupled to the inflaton via operators suppressed by a high mass scale Λ. We study a simple effective field theory in which a hidden sector field is coupled to a shift-symmetric inflaton via arbitrary operators up to dimension five. Self-interactions in the hidden sector lead to non-Gaussianity in the curvature perturbations. To be consistent with the Planck limit on local non-Gaussianity, the coupling to any hidden sector with light fields and natural cubic couplings must be suppressed by a very high scale Λ> 10^5 H. Even if the hidden sector has Gaussian correlations, nonlinearities in the mixing with the inflaton still lead to non-Gaussian curvature perturbations. In this case, the non-Gaussianity is of the equilateral or orthogonal type, and the Planck data requires Λ> 10^2 H. △ Less

Submitted 18 April, 2013; originally announced April 2013.

Comments: 40 pages, 11 figures

Symmetries and Loops in Inflation

Authors: Valentin Assassi, Daniel Baumann, Daniel Green

Abstract: In this paper, we prove that the superhorizon conservation of the curvature perturbation zeta in single-field inflation holds as an operator statement. This implies that all zeta-correlators are time independent at all orders in the loop expansion. Our result follows directly from locality and diffeomorphism invariance of the underlying theory. We also explore the relationship between the conserva… ▽ More In this paper, we prove that the superhorizon conservation of the curvature perturbation zeta in single-field inflation holds as an operator statement. This implies that all zeta-correlators are time independent at all orders in the loop expansion. Our result follows directly from locality and diffeomorphism invariance of the underlying theory. We also explore the relationship between the conservation of zeta, the single-field consistency relation and the renormalization of composite operators. △ Less

Submitted 8 November, 2012; v1 submitted 29 October, 2012; originally announced October 2012.

Comments: 31 pages, 1 figure; v2: references added + pedagogical improvements of Sec. 4

Report number: SITP 12/36

On Soft Limits of Inflationary Correlation Functions

Authors: Valentin Assassi, Daniel Baumann, Daniel Green

Abstract: Soft limits of inflationary correlation functions are both observationally relevant and theoretically robust. Various theorems can be proven about them that are insensitive to detailed model-building assumptions. In this paper, we re-derive several of these theorems in a universal way. Our method makes manifest why soft limits are such an interesting probe of the spectrum of additional light field… ▽ More Soft limits of inflationary correlation functions are both observationally relevant and theoretically robust. Various theorems can be proven about them that are insensitive to detailed model-building assumptions. In this paper, we re-derive several of these theorems in a universal way. Our method makes manifest why soft limits are such an interesting probe of the spectrum of additional light fields during inflation. We illustrate these abstract results with a detailed case study of the soft limits of quasi-single-field inflation. △ Less

Submitted 25 June, 2012; v1 submitted 18 April, 2012; originally announced April 2012.

Comments: 26 pages, 5 figures; V2: references added + pedagogical improvements of Sec. 2 and App. A