Tree search has become as a representative framework for test-time reasoning with large language models (LLMs), exemplified by methods such as Tree-of-Thought and Monte Carlo Tree Search. However, it remains difficult to provide instant and reliable quantitative assessments of intermediate reasoning step quality, and extensive path exploration is computationally costly. To address this, we propose Mutual Information Tree Search (MITS), a novel framework that guides reasoning with information-theoretic principles. MITS introduces an effective scoring function based on pointwise mutual information (PMI), which enables step-wise evaluation of reasoning paths and search tree expansion via beam search without expensive look-ahead simulations, achieving superior reasoning performances while maintaining computational efficiency. The framework is complemented by an entropy-based dynamic sampling strategy that adaptively allocates computational resources to uncertain reasoning steps where exploration is most beneficial. For final prediction, MITS employs a weighted voting scheme that combines PMI scores with prediction consensus. Through comprehensive experiments on diverse reasoning benchm
General-domain large multimodal models (LMMs) have achieved significant advances in various image-text tasks. However, their performance in the Intelligent Traffic Surveillance (ITS) domain remains limited due to the absence of dedicated multimodal datasets. To address this gap, we introduce MITS (Multimodal Intelligent Traffic Surveillance), the first large-scale multimodal benchmark dataset specifically designed for ITS. MITS includes 170,400 independently collected real-world ITS images sourced from traffic surveillance cameras, annotated with eight main categories and 24 subcategories of ITS-specific objects and events under diverse environmental conditions. Additionally, through a systematic data generation pipeline, we generate high-quality image captions and 5 million instruction-following visual question-answer pairs, addressing five critical ITS tasks: object and event recognition, object counting, object localization, background analysis, and event reasoning. To demonstrate MITS's effectiveness, we fine-tune mainstream LMMs on this dataset, enabling the development of ITS-specific applications. Experimental results show that MITS significantly improves LMM performance in
In the evolving landscape of quantum computing, determining the most efficient parameters for Quantum Error Correction (QEC) is paramount. Various quantum computers possess varied types and amounts of physical noise. Traditionally, simulators operate in a forward paradigm, taking parameters such as distance, rounds, and physical error to output a logical error rate. However, usage of maximum distance and rounds of the surface code might waste resources. An approach that relies on trial and error to fine-tune QEC code parameters using simulation tools like STIM can be exceedingly time-consuming. Additionally, daily fluctuations in quantum error rates can alter the ideal QEC settings needed. As a result, there is a crucial need for an automated solution that can rapidly determine the appropriate QEC parameters tailored to the current conditions. To bridge this gap, we present MITS, a tool designed to reverse-engineer the well-known simulator STIM for designing QEC codes. MITS accepts the specific noise model of a quantum computer and a target logical error rate as input and outputs the optimal surface code rounds and code distances. This guarantees minimal qubit and gate usage, harmo
The progress in generative models, particularly Generative Adversarial Networks (GANs), opened new possibilities for image generation but raised concerns about potential malicious uses, especially in sensitive areas like medical imaging. This study introduces MITS-GAN, a novel approach to prevent tampering in medical images, with a specific focus on CT scans. The approach disrupts the output of the attacker's CT-GAN architecture by introducing finely tuned perturbations that are imperceptible to the human eye. Specifically, the proposed approach involves the introduction of appropriate Gaussian noise to the input as a protective measure against various attacks. Our method aims to enhance tamper resistance, comparing favorably to existing techniques. Experimental results on a CT scan demonstrate MITS-GAN's superior performance, emphasizing its ability to generate tamper-resistant images with negligible artifacts. As image tampering in medical domains poses life-threatening risks, our proactive approach contributes to the responsible and ethical use of generative models. This work provides a foundation for future research in countering cyber threats in medical imaging. Models and cod
We highlight a simple congruence transform that shifts coupling parameters for Dirac operators with shell interactions. As one of the consequences, this leads to new observations concerning the self-adjointness and the infinite mass interpretation of generalized MIT bag models.
This paper presents an improved technique for solving the inverse problem in magnetic induction tomography (MIT) by considering skin and proximity effects in coils. MIT is a non-contact, noninvasive, and low-cost imaging modality for obtaining the distribution of conductivity inside an object. Reconstruction of low conductivity distribution by MIT requires more accurate techniques since measured signals are inherently weak and the reconstruction problem is highly nonlinear and ill-posed. Previous MIT inverse problem studies have ignored skin and proximity effects inside coils in the forward method. In this article, the improved technique incorporates these effects in the forward method. Furthermore, it employs the regularized Gauss-Newton algorithm to reconstruct the conductivity distribution. The regularization parameter is obtained by an adaptive method using the two input parameters: a coefficient and an initial conductivity distribution. The new Jacobian matrix is computed based on a standard technique. To compare the early and improved forward methods in possible medical and industrial applications with low conductivity regions, a 2D 8-coil MIT system is modeled, and image rec
The Robin criterion states that the Riemann hypothesis is equivalent to the inequality $σ(n) < e^γn \log \log n$ for all $n>5040$, where $σ(n)$ is the sum of divisors of $n$, and $γ$ is the Euler--Mascheroni constant. Define the family of functions \[ σ^{[k]} (n):=\sum_{[d_1,\dots,d_k]=n}d_1\dots d_k \] where $[d_1, \dots, d_k]$ is the least common multiple of $d_1, \dots, d_k$. These functions behave asymptotically like $σ(n)^k$ as $k\to\infty$. We prove the following analogue of the Robin criterion: for any $k \geq 2$, the Riemann hypothesis holds if and only if $σ^{[k]} (n) < \frac{(e^γn \log \log n)^k}{ζ(k)}$ for all $n > 2162160$, where $ζ$ is the Riemann zeta function.
The Dirac operator with MIT bag boundary condition in a bounded convex domain is shown to be always self-adjoint in the $H^1$-setting. This allows one to show that such operators appear as limit of Dirac operators with large positive mass outside the domain. Similar results were previously known for smooth domains only.
In this document, we examine exact and efficient numerical approaches to the MIT Bag Model, a theoretical framework used to describe the properties of bound quarks in Hadrons. We present the exact and Boundary Value Problem (BVP) numerical approaches. Both methods are effective in calculating the eigen-functions and energy levels. Notably, the precision of the BVP approach matches up to 10 decimal places when compared to the exact approach.
Image Translation (IT) holds immense potential across diverse domains, enabling the translation of textual content within images into various languages. However, existing datasets often suffer from limitations in scale, diversity, and quality, hindering the development and evaluation of IT models. To address this issue, we introduce MIT-10M, a large-scale parallel corpus of multilingual image translation with over 10M image-text pairs derived from real-world data, which has undergone extensive data cleaning and multilingual translation validation. It contains 840K images in three sizes, 28 categories, tasks with three levels of difficulty and 14 languages image-text pairs, which is a considerable improvement on existing datasets. We conduct extensive experiments to evaluate and train models on MIT-10M. The experimental results clearly indicate that our dataset has higher adaptability when it comes to evaluating the performance of the models in tackling challenging and complex image translation tasks in the real world. Moreover, the performance of the model fine-tuned with MIT-10M has tripled compared to the baseline model, further confirming its superiority.
In the present work we use the vector MIT bag model to describe quark matter and obtain the macroscopic properties of neutron stars. We also calculate the speed of sound of quark matter described by this model with specific parameter values and check if the results obey the conformal limit at high densities. We have seen that the vector MIT bag model produces stars with masses and radii consistent with recent data, suggesting the possible existence of massive quark (or strange) stars.
We curate a comprehensive dataset of 4,550 questions and solutions from problem sets, midterm exams, and final exams across all MIT Mathematics and Electrical Engineering and Computer Science (EECS) courses required for obtaining a degree. We evaluate the ability of large language models to fulfill the graduation requirements for any MIT major in Mathematics and EECS. Our results demonstrate that GPT-3.5 successfully solves a third of the entire MIT curriculum, while GPT-4, with prompt engineering, achieves a perfect solve rate on a test set excluding questions based on images. We fine-tune an open-source large language model on this dataset. We employ GPT-4 to automatically grade model responses, providing a detailed performance breakdown by course, question, and answer type. By embedding questions in a low-dimensional space, we explore the relationships between questions, topics, and classes and discover which questions and classes are required for solving other questions and classes through few-shot learning. Our analysis offers valuable insights into course prerequisites and curriculum design, highlighting language models' potential for learning and improving Mathematics and EE
The purpose of this paper is to introduce and study Poincaré-Steklov (PS) operators associated to the Dirac operator $D_m$ with the so-called MIT bag boundary condition. In a domain $Ω\subset\mathbb{R}^3$, for a complex number $z$ and for $U_z$ a solution of $(D_m-z)U_z=0$, the associated PS operator maps the value of $Γ_- U_z$, the MIT bag boundary value of $U_z$, to $Γ_+ U_z$, where $Γ_\pm$ are projections along the boundary $\partialΩ$ and $(Γ_ - + Γ_+) = t_{\partialΩ}$ is the trace operator on $\partialΩ$. In the first part of this paper, we show that the PS operator is a zero-order pseudodifferential operator and give its principal symbol. In the second part, we study the PS operator when the mass $m$ is large, and we prove that it fits into the framework of $1/m$-pseudodifferential operators, and we derive some important properties, especially its semiclassical principal symbol. Subsequently, we apply these results to establish a Krein-type resolvent formula for the Dirac operator $H_M= D_m+ Mβ1_{\mathbb{R}^3\setminus\overlineΩ}$ for large masses $M>0$, in terms of the resolvent of the MIT bag operator on $Ω$. With its help, the large coupling convergence with a convergenc
The goal of this report is to provide actionable recommendations for paths to carbon neutrality from MITs largest group of stakeholders, students. In the report, we consider the optimization of emissions reductions, the timescale of solution implementation, cost, scalability, and the associated risks inherent to two different scenarios. Similarly, it leverages MITs strengths, such as its position as the world leader in technology and innovation as well as its financial strength, which enables it to make relatively low return investments and use contracting power to enable renewable projects.
This RFC white Paper summarizes our progress on the MIT Voice Name System (VNS) and Huey. The VNS, similar in name and function to the DNS, is a system to reserve and use "wake words" to activate Artificial Intelligence (AI) devices. Just like you can say "Hey Siri" to activate Apple's personal assistant, we propose using the VNS in smart speakers and other devices to route wake requests based on commands such as "turn off", "open grocery shopping list" or "271, start flash card review of my computer vision class". We also introduce Huey, an unambiguous Natural Language to interact with AI devices. We aim to standardize voice interactions to a universal reach similar to that of other systems such as phone numbering, with an agreed world-wide approach to assign and use numbers, or the Internet's DNS, with a standard naming system, that has helped flourish popular services including the World-Wide-Web, FTP, and email. Just like these standards are "neutral", we also aim to endow the VNS with "wake neutrality" so that each participant can develop its own digital voice. We focus on voice as a starting point to talk to any IoT object and explain briefly how the VNS may be expanded to ot
High-Performance Computing (HPC) centers and cloud providers support an increasingly diverse set of applications on heterogenous hardware. As Artificial Intelligence (AI) and Machine Learning (ML) workloads have become an increasingly larger share of the compute workloads, new approaches to optimized resource usage, allocation, and deployment of new AI frameworks are needed. By identifying compute workloads and their utilization characteristics, HPC systems may be able to better match available resources with the application demand. By leveraging datacenter instrumentation, it may be possible to develop AI-based approaches that can identify workloads and provide feedback to researchers and datacenter operators for improving operational efficiency. To enable this research, we released the MIT Supercloud Dataset, which provides detailed monitoring logs from the MIT Supercloud cluster. This dataset includes CPU and GPU usage by jobs, memory usage, and file system logs. In this paper, we present a workload classification challenge based on this dataset. We introduce a labelled dataset that can be used to develop new approaches to workload classification and present initial results base
For a family of self-adjoint Dirac operators $-i c (α\cdot abla) + \frac{c^2}{2}$ subject to generalized MIT bag boundary conditions on domains in $\mathbb R^3$ it is shown that the nonrelativistic limit in the norm resolvent sense is the Dirichlet Laplacian. This allows to transfer spectral geometry results for Dirichlet Laplacians to Dirac operators for large $c$.
Recently V.H.López Solís and I.Shestakov solved an old problem by N.Jacobson on describing of unital alternative algebras containing the $2\times 2$ matrix algebra $M_2$ as a unital subalgebra. Here we give another description of $M_2$-algebras via the 6-dimensional alternative superalgebra $B(4,2)$ and an auxiliar $\mathbf {\mit Z}_2$-graded algebra $\mitΓ$. It occurs that the category of alternative $M_2$-algebras is isomorphic to the category of $\mitΓ$-algebras. For any associative and commutative algebra $A$, we give a construction of a ${\mit Γ}$-algebra $\mitΓ(A)$, which turns to be a Jordan superalgebra; if $A$ is a domain then $\mitΓ(A)$ is a prime superalgebra. We describe also the free $\mitΓ$-algebras and construct their bases.
Ultrafast X-rays revealed how a molecule converts absorbed light into motion in just trillionths of a second。 Individual atoms recorded different stages of the process, opening a powerful new window into light-driven chemistry