Search

Recent advancements in large-scale multi-task robot learning offer the potential for deploying robot fleets in household and industrial settings, enabling them to perform diverse tasks across various environments. However, AI-enabled robots often face challenges with generalization and robustness when exposed to real-world variability and uncertainty. We introduce Sirius-Fleet, a multi-task interactive robot fleet learning framework to address these challenges. Sirius-Fleet monitors robot performance during deployment and involves humans to correct the robot's actions when necessary. We employ a visual world model to predict the outcomes of future actions and build anomaly predictors to predict whether they will likely result in anomalies. As the robot autonomy improves, the anomaly predictors automatically adapt their prediction criteria, leading to fewer requests for human intervention and gradually reducing human workload over time. Evaluations on large-scale benchmarks demonstrate Sirius-Fleet's effectiveness in improving multi-task policy performance and monitoring accuracy. We demonstrate Sirius-Fleet's performance in both RoboCasa in simulation and Mutex in the real world, two diverse, large-scale multi-task benchmarks. More information is available on the project website: https://ut-austin-rpl.github.io/sirius-fleet, Comment: In Proceedings of CoRL 2024

Activation sparsity can enable practical inference speedups in large language models (LLMs) by reducing the compute and memory-movement required for matrix multiplications during the forward pass. However, existing methods face limitations that inhibit widespread adoption. Some approaches are tailored towards older models with ReLU-based sparsity, while others require extensive continued pre-training on up to hundreds of billions of tokens. This paper describes TEAL, a simple training-free method that applies magnitude-based activation sparsity to hidden states throughout the entire model. TEAL achieves 40-50% model-wide sparsity with minimal performance degradation across Llama-2, Llama-3, and Mistral families, with sizes varying from 7B to 70B. We improve existing sparse kernels and demonstrate wall-clock decoding speed-ups of up to 1.53$\times$ and 1.8$\times$ at 40% and 50% model-wide sparsity. TEAL is compatible with weight quantization, enabling further efficiency gains., Comment: Rev. 1: fixing minor typos

There is now strong evidence that the superconformal index of holographic gauge theories admits a giant graviton expansion capturing the worldvolume dynamics of spherical D3-branes moving in internal space. While the giant graviton indices may be written as suitable contour integrals, care must be taken in their evaluation, as there can be an ambiguity in choice of integration contour or, equivalently, "pole selection". We resolve this ambiguity by using Murthy's recent expansion of the superconformal index to provide a rigorous underpinning of the evaluation of the supergravity giant graviton indices. In doing so, we directly relate Murthy's giant graviton expansion to the supergravity D3-brane expansion., Comment: 31 pages

Ignoring ten-dimensional heterotic gauge fields, heterotic supergravity reduced on a $d$-dimensional torus gives rise to a half-maximal supergravity coupled to $d$ vector multiplets. The reduced theory has a continuous $O(d,d;\mathbb R)/O(d)_-\times O(d)_+$ symmetry that persists to all perturbative orders in the string $\alpha'$ expansion. We highlight this symmetry by explicitly reducing the bosonic sector of four-derivative heterotic supergravity as well as its fermionic supersymmetry variations. After appropriate field redefinitions, the resulting action and supersymmetry variations are manifestly $O(d)_-\times O(d)_+$ invariant. This reduction allows us to explore the interplay between the gravity and vector multiplets beyond leading order, where (in our conventions) $O(d)_-$ is the supergravity R-symmetry while $O(d)_+$ is a flavor symmetry of the $d$ vector multiplets., Comment: 26 pages, references added

We revisit the existence of monotonic quantities along renormalization group flows using only the Null Energy Condition and the Ryu-Takayanagi formula for the entanglement entropy of field theories with anti-de Sitter gravity duals. In particular, we consider flows within the same dimension and holographically reprove the $c$-, $F$-, and $a$-theorems in dimensions two, three, and four. We focus on the family of maximally spherical entangling surfaces, define a quasi-constant of motion corresponding to the breaking of conformal invariance, and use a properly defined distance between minimal surfaces to construct a holographic $c$-function that is monotonic along the flow. We then apply our method to the case of flows across dimensions: There, we reprove the monotonicity of flows from $\mathrm{AdS}_{D+1}$ to $\mathrm{AdS}_3$ and prove the novel case of flows from $\mathrm{AdS}_5$ to $\mathrm{AdS}_4$., Comment: 28 pages, 3 figures; references added

The superconformal index of half-BPS states in ${\cal N}=4$ supersymmetric Yang-Mills with gauge group $U(N)$ admits an expansion in terms of giant gravitons, ${\cal I}_N(q)={\cal I}_\infty(q) \sum\limits_{m=0}^\infty q^{mN}\hat{\mathcal I}_m(q)$, where $m$ is the number of giant gravitons. We derive this expansion directly in supergravity from the class of half-BPS solutions due to Lin, Lunin, and Maldacena in type IIB supergravity. The moduli space of these configurations can be quantized using covariant quantization methods. We review how this quantization leads to the graviton index, ${\cal I}_\infty(q)$, and present a modification that leads to the precise expression for the expansion in terms of giant gravitons. Our proposal provides a derivation of the giant graviton expansion directly in terms of supergravity degrees of freedom. We also comment on how to derive the expansion in terms of the effective Fermi droplet picture., Comment: 17 pages, 4 figures

Large Language Models (LLMs) are typically trained in two phases: pre-training on large internet-scale datasets, and fine-tuning for downstream tasks. Given the higher computational demand of pre-training, it's intuitive to assume that fine-tuning adds less new information to the model, and is thus more compressible. We explore this assumption by decomposing the weights of fine-tuned models into their pre-trained components and an additional delta. We introduce a simple method, BitDelta, which successfully quantizes this delta down to 1 bit without compromising performance. This interesting finding not only highlights the potential redundancy of information added during fine-tuning, but also has significant implications for the multi-tenant serving and multi-tenant storage of fine-tuned models. By enabling the use of a single high-precision base model accompanied by multiple 1-bit deltas, BitDelta dramatically reduces GPU memory requirements by more than 10x, which can also be translated to enhanced generation latency in multi-tenant settings. We validate BitDelta through experiments across Llama-2 and Mistral model families, and on models up to 70B parameters, showcasing minimal performance degradation over all tested settings., Comment: NeurIPS 2024 acceptance

We study the entanglement entropy associated with a holographic RG flow from $\textrm{AdS}_7$ to $\textrm{AdS}_{4} \times \mathbb{H}_3$, where $\mathbb{H}_3$ is a $3$-dimensional hyperbolic manifold with curvature $\kappa$. The dual six-dimensional RG flow is disconnected from Lorentz-invariant flows. In this context we address various notions of central charges and identify a monotonic candidate $c$-function that captures IR aspects of the flow. The UV behavior of the holographic entanglement entropy and, in particular its universal term, display an interesting dependence on the curvature, $\kappa$. We then contrast our holographic results with existing field theory computations in six dimensions and find a series of new corrections in curvature to the universal term in the entanglement entropy., Comment: 31 pages, 9 figures. v2: References added, matches JHEP version

At the two-derivative order, the group manifold reduction of heterotic supergravity on $S^3$ results in a half-maximal 7D gauged supergravity coupled to three vector multiplets, and a further truncation can be taken to remove the vector multiplets. We demonstrate that this truncation remains consistent at the four-derivative level; we do so both by analysis of the equations of motion and the supersymmetry variations., Comment: 21 pages; references added

We consider the torus reduction of heterotic supergravity in the presence of four-derivative corrections. In particular, the reduction on $T^n$ generically leads to a half-maximal supergravity coupled to $n$ vector multiplets, and we show that it is consistent to truncate out said vector multiplets. This is done by the analysis of both the bosonic equations of motion and the Killing spinor equations. As an application of the consistent truncation, we examine the four-derivative corrected BPS black string that reduces to a black hole in minimal nine-dimensional supergravity., Comment: 34 pages; references added

In the context of gravitational theories describing renormalization group flows across dimensions via AdS/CFT, we study the role of higher-derivative corrections to Einstein gravity. We use the Null Energy Condition to derive monotonicity properties of candidate holographic central charges formed by combinations of metric functions. We also implement an entropic approach to the characterization of the four-derivative flows using the Jacobson-Myers functional and demonstrate, under reasonable conditions, monotonicity of certain terms in the entanglement entropy via the appropriate generalization of the Ryu-Takayanagi prescription. In particular, we show that any flow from a higher dimensional theory to a holographic CFT$_2$ satisfies a type of monotonicity. We also uncover direct relations between NEC-motivated and entropic central charges., Comment: 37 pages; references added

The superconformal index of $\mathcal N=4$ super-Yang Mills theory with $U(N)$ gauge group can be written as a matrix integral over the gauge group. Recently, Murthy demonstrated that this integral can be reexpressed as a sum of terms corresponding to a giant graviton expansion of the index, and provided an explicit formula for the case of a single giant graviton. Here we give similar explicit formulae for an arbitrary number, $m\ge1$, of giant gravitons. We provide 1/2 and 1/16 BPS index examples up to the order where three giant gravitons contribute and demonstrate that the expansion of the matrix integral differs from the giant graviton expansion computed in the supergravity dual. This shows that the giant graviton expansion is not necessarily unique once two or more giant gravitons start appearing., Comment: 24 pages, no figures, reference added

T-duality has been shown to constrain the higher derivative corrections of string theory. We revisit the problem of understanding the T-duality constraints imposed on the $\alpha'$ corrections using the language of a torsionful connection. We find a convenient way to express the $O(d,d)$ invariants in terms of linear combinations of the metric and B-field for general $d$-dimensional torus compactifications, which we then use to revisit the heterotic and bosonic ten-dimensional string action at order $\alpha'$. We also comment on the four-point functions for corrections of order $\alpha'^3$ in the type II string.

We present a comprehensive analysis of the evolution of the morphological and structural properties of a large sample of galaxies at z=3-9 using early JWST CEERS NIRCam observations. Our sample consists of 850 galaxies at z>3 detected in both CANDELS HST imaging and JWST CEERS NIRCam images to enable a comparison of HST and JWST morphologies. Our team conducted a set of visual classifications, with each galaxy in the sample classified by three different individuals. We also measure quantitative morphologies using the publicly available codes across all seven NIRCam filters. Using these measurements, we present the fraction of galaxies of each morphological type as a function of redshift. Overall, we find that galaxies at z>3 have a wide diversity of morphologies. Galaxies with disks make up a total of 60\% of galaxies at z=3 and this fraction drops to ~30% at z=6-9, while galaxies with spheroids make up ~30-40% across the whole redshift range and pure spheroids with no evidence for disks or irregular features make up ~20%. The fraction of galaxies with irregular features is roughly constant at all redshifts (~40-50%), while those that are purely irregular increases from ~12% to ~20% at z>4.5. We note that these are apparent fractions as many selection effects impact the visibility of morphological features at high redshift. The distributions of S\'ersic index, size, and axis ratios show significant differences between the morphological groups. Spheroid Only galaxies have a higher S\'ersic index, smaller size, and higher axis ratio than Disk/Irregular galaxies. Across all redshifts, smaller spheroid and disk galaxies tend to be rounder. Overall, these trends suggest that galaxies with established disks and spheroids exist across the full redshift range of this study and further work with large samples at higher redshift is needed to quantify when these features first formed., Comment: Accepted for publication to ApJL, 24 pages, 14 figures

We explore the notion of $c$-functions in renormalization group flows between theories in different spacetime dimensions. We discuss functions connecting central charges of the UV and IR fixed point theories on the one hand, and functions which are monotonic along the flow on the other. First, using the geometric properties of the holographic dual RG flows across dimensions and the constraints from the null energy condition, we construct a monotonic holographic $c$-function and thereby establish a holographic $c$-theorem across dimensions. Second, we use entanglement entropies for two different types of entangling regions in a field theory along the RG flow across dimensions to construct candidate $c$-functions which satisfy one of the two criteria but not both. In due process we also discuss an interesting connection between corner contributions to the entanglement entropy and the topology of the compact internal space. As concrete examples for both approaches, we holographically study twisted compactifications of 4d $\mathcal N=4$ SYM and compactifications of 6d $\mathcal N=(2,0)$ theories., Comment: 49 pages, 17 figures; v2: minor corrections

We find a class of approximate perturbative solutions describing multi-centered BPS black holes in asymptotically AdS$_4$. These black holes are moving coplanarly in a circular orbit with common radius but fixed phase differences, where the radius is characterized by a common boost velocity $v$. While the solutions are not complete, as they are only accurate up to $\mathcal{O}(M^2v^2)$, we are able to locally investigate their supersymmetric properties in four-dimensional $\mathcal{N} = 2$ supergravity. The result indicates that, while a single BPS black hole preserves half of the supersymmetries, a configuration with two or more black holes is only 1/4 BPS. These perturbative solutions reduce in the asymptotically Minkowski limit to the multi-black hole solution of Majumdar and Papapetrou., Comment: 27 pages

We study the non-linear structure of Type IIB eight-derivative couplings involving the metric and the complexified three-form $G_3$. We show that, at the level of five-point string amplitudes, the kinematics in the maximally R-symmetry-violating sector is fully matched by standard superspace integrals and by superparticle amplitudes in M-theory on a two-torus. The latter approach is used to determine the complete effective action in this sector and to verify its invariance under SL$(2,\mathbb{Z})$ duality. We further comment on the general structure of the higher-point kinematics. Compactifications to lower dimensions provide both tests for our results and the arena for their applications. We verify that K3 reductions are fully consistent with the constraints of six-dimensional supersymmetry, and derive the four-dimensional flux scalar potential and axion kinetic terms at order $(\alpha^{\prime})^{3}$ in Calabi-Yau threefold reductions., Comment: 34 pages plus appendices