Search

Information retrieval, specifically contract clause retrieval, is foundational to contract drafting because lawyers rarely draft contracts from scratch; instead, they locate and revise the most relevant precedent. We introduce the Atticus Clause Retrieval Dataset (ACORD), the first retrieval benchmark for contract drafting fully annotated by experts. ACORD focuses on complex contract clauses such as Limitation of Liability, Indemnification, Change of Control, and Most Favored Nation. It includes 114 queries and over 126,000 query-clause pairs, each ranked on a scale from 1 to 5 stars. The task is to find the most relevant precedent clauses to a query. The bi-encoder retriever paired with pointwise LLMs re-rankers shows promising results. However, substantial improvements are still needed to effectively manage the complex legal work typically undertaken by lawyers. As the first retrieval benchmark for contract drafting annotated by experts, ACORD can serve as a valuable IR benchmark for the NLP community.

Our work examines the efficacy of employing advanced machine learning methods to solve captchas from Google's reCAPTCHAv2 system. We evaluate the effectiveness of automated systems in solving captchas by utilizing advanced YOLO models for image segmentation and classification. Our main result is that we can solve 100% of the captchas, while previous work only solved 68-71%. Furthermore, our findings suggest that there is no significant difference in the number of challenges humans and bots must solve to pass the captchas in reCAPTCHAv2. This implies that current AI technologies can exploit advanced image-based captchas. We also look under the hood of reCAPTCHAv2, and find evidence that reCAPTCHAv2 is heavily based on cookie and browser history data when evaluating whether a user is human or not. The code is provided alongside this paper., Comment: 10 pages. Accepted at COMPSAC 2024

Bessel functions are critical in scientific computing for applications such as machine learning, protein structure modeling, and robotics. However, currently, available routines lack precision or fail for certain input ranges, such as when the order $v$ is large, and GPU-specific implementations are limited. We address the precision limitations of current numerical implementations while dramatically improving the runtime. We propose two novel algorithms for computing the logarithm of modified Bessel functions of the first and second kinds by computing intermediate values on a logarithmic scale. Our algorithms are robust and never have issues with underflows or overflows while having relative errors on the order of machine precision, even for inputs where existing libraries fail. In C++/CUDA, our algorithms have median and maximum speedups of 45x and 6150x for GPU and 17x and 3403x for CPU, respectively, over the ranges of inputs and third-party libraries tested. Compared to SciPy, the algorithms have median and maximum speedups of 77x and 300x for GPU and 35x and 98x for CPU, respectively, over the tested inputs. The ability to robustly compute a solution and the low relative errors allow us to fit von Mises-Fisher, vMF, distributions to high-dimensional neural network features. This is, e.g., relevant for uncertainty quantification in metric learning. We obtain image feature data by processing CIFAR10 training images with the convolutional layers of a pre-trained ResNet50. We successfully fit vMF distributions to 2048-, 8192-, and 32768-dimensional image feature data using our algorithms. Our approach provides fast and accurate results while existing implementations in SciPy and mpmath fail to fit successfully. Our approach is readily implementable on GPUs, and we provide a fast open-source implementation alongside this paper., Comment: Accepted at ICS 2024

This paper presents SYBILGAT, a novel approach to Sybil detection in social networks using Graph Attention Networks (GATs). Traditional methods for Sybil detection primarily leverage structural properties of networks; however, they tend to struggle with a large number of attack edges and are often unable to simultaneously utilize both known Sybil and honest nodes. Our proposed method addresses these limitations by dynamically assigning attention weights to different nodes during aggregations, enhancing detection performance. We conducted extensive experiments in various scenarios, including pretraining in sampled subgraphs, synthetic networks, and networks under targeted attacks. The results show that SYBILGAT significantly outperforms the state-of-the-art algorithms, particularly in scenarios with high attack complexity and when the number of attack edges increases. Our approach shows robust performance across different network models and sizes, even as the detection task becomes more challenging. We successfully applied the model to a real-world Twitter graph with more than 269k nodes and 6.8M edges. The flexibility and generalizability of SYBILGAT make it a promising tool to defend against Sybil attacks in online social networks with only structural information., Comment: 9 pages, 1 figure, 6 tables

Modern CAPTCHAs rely heavily on vision tasks that are supposedly hard for computers but easy for humans. However, advances in image recognition models pose a significant threat to such CAPTCHAs. These models can easily be fooled by generating some well-hidden "random" noise and adding it to the image, or hiding objects in the image. However, these methods are model-specific and thus can not aid CAPTCHAs in fooling all models. We show in this work that by allowing for more significant changes to the images while preserving the semantic information and keeping it solvable by humans, we can fool many state-of-the-art models. Specifically, we demonstrate that by adding masks of various intensities the Accuracy @ 1 (Acc@1) drops by more than 50%-points for all models, and supposedly robust models such as vision transformers see an Acc@1 drop of 80%-points. These masks can therefore effectively fool modern image classifiers, thus showing that machines have not caught up with humans -- yet., Comment: Under review

We construct homomorphic images of $su(n,n)^{\mathbb C}$ in Weyl Algebras ${\mathcal H}_{2nr}$. More precisely, and using the Bernstein filtration of ${\mathcal H}_{2nr}$, $su(n,n)^{\mathbb C}$ is mapped into degree $2$ elements with the negative non-compact root spaces being mapped into second order creation operators. Using the Fock representation of ${\mathcal H}_{2nr}$, these homomorphisms give all unitary highest weight representations of $su(n,n)^{\mathbb C}$ thus reconstructing the Kashiwara--Vergne List for the Segal--Shale--Weil representation. Just as in the derivation of the their list, we construct a representation of $u(r)$ in the Fock space commuting with $su(n,n)^{\mathbb C}$, and this gives the multiplicities. The construction also gives an easy proof that the ideals of $(r+1)\times (r+1)$ minors are prime ($r\leq n-1)$. The quotients of all polynomials by such ideals carry the more singular of the representations. As a consequence, these representations can be realized in spaces of solutions to Maxwell type equations. We actually go one step further and determine exactly which representations from our list are missing some ${\mathfrak k}^{\mathbb C}$-types, thereby revealing exactly which covariant differential operators have unitary null spaces. We prove the analogous results for ${\mathcal U}_q(su(n,n)^{\mathbb C})$. The Weyl Algebras are replaced by the Hayashi--Weyl Algebras ${\mathcal H}{\mathcal W}_{2nr}$ and the Fock space by a $q$-Fock space. Further, determinants are replaced by $q$-determinants, and a commuting representation of ${\mathcal U}_q(u(r))$ in the $q$-Fock space is constructed. For this purpose a Drinfeld Double is used. We mention one difference: The quantized negative non-compact root spaces, while still of degree 2, are no longer given entirely by second order creation operators., Comment: 66 pages LaTeX

Are low-income individuals relying on government transfers liquidity constrained by the end of the month to a degree that they postpone medical treatment? I investigate this question using Danish administrative data comprising the universe of welfare recipients and the filling of all prescription drugs. I find that on transfer income payday, recipients have a 52% increase in the propensity to fill a prescription. By separating prophylaxis drugs used to treat chronic conditions, where the patient can anticipate the need to fill the prescription, e.g. cholesterol-lowering statins, I find an increase of up to 99% increase on payday. Even for drugs used to treat acute conditions, where timely treatment is essential, I find a 22% increase on payday for antibiotics and a 5-8% decrease in the four days preceding payday. Lastly, exploiting the difference in day the doctor write the prescription and the day the patient fill it, I show that liquidity constraints is the key operating mechanism for postponing antibiotic treatment.

We study homomorphisms between quantized generalized Verma modules $M(V_{\Lambda})\stackrel{\phi_{\Lambda,\Lambda_1}}{\rightarrow}M(V_{\Lambda_1})$ for ${\mathcal U}_q(su(n,n))$. There is a natural notion of degree for such maps, and if the map is of degree $k$, we write $\phi^k_{\Lambda,\Lambda_1}$. We examine when one can have a series of such homomorphisms $\phi^1_{\Lambda_{n-1},\Lambda_{n}} \circ \phi^1_{\Lambda_{n-2}, \Lambda_{n-1}} \circ\cdots\circ \phi^1_{\Lambda,\Lambda_1} = \textrm{Det}_q$, where $\textrm{Det}_q$ denotes the map $M(V_{\Lambda})\ni p\rightarrow \textrm{Det}_q\cdot p\in M(V_{\Lambda_n})$. If, classically, $su(n,n)^{\mathbb C}={\mathfrak p}^-\oplus(su(n)\oplus su(n)\oplus {\mathbb C})\oplus {\mathfrak p}^+$, then $\Lambda = (\Lambda_L,\Lambda_R,\lambda)$ and $\Lambda_n =(\Lambda_L,\Lambda_R,\lambda+2)$. The answer is then that $\Lambda$ must be one-sided in the sense that either $\Lambda_L=0$ or $\Lambda_R=0$ (non-exclusively). There are further demands on $\lambda$ if we insist on ${\mathcal U}_q({\mathfrak g}^{\mathbb C})$ homomorphisms. However, it is also interesting to loosen this to considering only ${\mathcal U}^-_q({\mathfrak g}^{\mathbb C})$ homomorphisms, in which case the conditions on $\lambda$ disappear. By duality, there result have implications on covariant quantized differential operators. We finish by giving an explicit, though sketched, determination of the full set of ${\mathcal U}_q({\mathfrak g}^{\mathbb C})$ homomorphisms $\phi^1_{\Lambda,\Lambda_1}$., Comment: 10 pages proceedings of Group 32, Prague 2018

In the framework of (vector valued) quantized holomorphic functions defined on non-commutative spaces, ``quantized hermitian symmetric spaces'', we analyze what the algebras of quantized differential operators with variable coefficients should be. It is an emediate point that even $0$th order operators, given as multiplications by polynomials, have to be specified as e.g. left or right multiplication operators since the polynomial algebras are replaced by quadratic, non-commutative algebras. In the settings we are interested in, there are bilinear pairings which allows us to define differential operators as duals of multiplication operators. Indeed, there are different choices of pairings which lead to quite different results. We consider three different pairings. The pairings are between quantized generalized Verma modules and quantized holomorphically induced modules. It is a natural demand that the corresponding representations can be expressed by (matrix valued) differential operators. We show that a quantum Weyl algebra ${\mathcal W}eyl_q(n,n)$ introduced by T. Hyashi (Comm. Math. Phys. 1990) plays a fundamental role. In fact, for one pairing, the algebra of differential operators, though inherently depending on a choice of basis, is precisely matrices over ${\mathcal W}eyl_q(n,n)$. We determine explicitly the form of the (quantum) holomorphically induced representations and determine, for the different pairings, if they can be expressed by differential operators., Comment: 37 pages LaTeX