Search

Generative AI holds the promise of enabling a range of sought-after capabilities and revolutionizing workflows in various consumer and enterprise verticals. However, putting a model in production involves much more than just generating an output. It involves ensuring the model is reliable, safe, performant and also adheres to the policy of operation in a particular domain. Guardrails as a necessity for models has evolved around the need to enforce appropriate behavior of models, especially when they are in production. In this paper, we use education as a use case, given its stringent requirements of the appropriateness of content in the domain, to demonstrate how a guardrail model can be trained and deployed in production. Specifically, we describe our experience in building a production-grade guardrail model for a K-12 educational platform. We begin by formulating the requirements for deployment to this sensitive domain. We then describe the training and benchmarking of our domain-specific guardrail model, which outperforms competing open- and closed- instruction-tuned models of similar and larger size, on proprietary education-related benchmarks and public benchmarks related to general aspects of safety. Finally, we detail the choices we made on architecture and the optimizations for deploying this service in production; these range across the stack from the hardware infrastructure to the serving layer to language model inference optimizations. We hope this paper will be instructive to other practitioners looking to create production-grade domain-specific services based on generative AI and large language models.

There has been significant recent interest in harnessing LLMs to control software systems through multi-step reasoning, planning and tool-usage. While some promising results have been obtained, application to specific domains raises several general issues including the control of specialized domain tools, the lack of existing datasets for training and evaluation, and the non-triviality of automated system evaluation and improvement. In this paper, we present a case-study where we examine these issues in the context of a specific domain. Specifically, we present an automated math visualizer and solver system for mathematical pedagogy. The system orchestrates mathematical solvers and math graphing tools to produce accurate visualizations from simple natural language commands. We describe the creation of specialized data-sets, and also develop an auto-evaluator to easily evaluate the outputs of our system by comparing them to ground-truth expressions. We have open sourced the data-sets and code for the proposed system.

AI Agents are changing the way work gets done, both in consumer and enterprise domains. However, the design patterns and architectures to build highly capable agents or multi-agent systems are still developing, and the understanding of the implication of various design choices and algorithms is still evolving. In this paper, we present our work on building a novel web agent, Agent-E \footnote{Our code is available at \url{https://github.com/EmergenceAI/Agent-E}}. Agent-E introduces numerous architectural improvements over prior state-of-the-art web agents such as hierarchical architecture, flexible DOM distillation and denoising method, and the concept of \textit{change observation} to guide the agent towards more accurate performance. We first present the results of an evaluation of Agent-E on WebVoyager benchmark dataset and show that Agent-E beats other SOTA text and multi-modal web agents on this benchmark in most categories by 10-30\%. We then synthesize our learnings from the development of Agent-E into general design principles for developing agentic systems. These include the use of domain-specific primitive skills, the importance of distillation and de-noising of environmental observations, the advantages of a hierarchical architecture, and the role of agentic self-improvement to enhance agent efficiency and efficacy as the agent gathers experience.

The use of question-based activities (QBAs) is wide-spread in education, traditionally forming an integral part of the learning and assessment process. In this paper, we design and evaluate an automated question generation tool for formative and summative assessment in schools. We present an expert survey of one hundred and four teachers, demonstrating the need for automated generation of QBAs, as a tool that can significantly reduce the workload of teachers and facilitate personalized learning experiences. Leveraging the recent advancements in generative AI, we then present a modular framework employing transformer based language models for automatic generation of multiple-choice questions (MCQs) from textual content. The presented solution, with distinct modules for question generation, correct answer prediction, and distractor formulation, enables us to evaluate different language models and generation techniques. Finally, we perform an extensive quantitative and qualitative evaluation, demonstrating trade-offs in the use of different techniques and models., Comment: Accepted to AI-2023 (Forty-third SGAI International Conference on Artificial Intelligence) as a long paper, link: http://www.bcs-sgai.org/ai2023

We present a Reinforcement Learning (RL) based framework for optimizing long-term discounted reward problems with large combinatorial action space and state dependent constraints. These characteristics are common to many operations management problems, e.g., network inventory replenishment, where managers have to deal with uncertain demand, lost sales, and capacity constraints that results in more complex feasible action spaces. Our proposed Programmable Actor Reinforcement Learning (PARL) uses a deep-policy iteration method that leverages neural networks (NNs) to approximate the value function and combines it with mathematical programming (MP) and sample average approximation (SAA) to solve the per-step-action optimally while accounting for combinatorial action spaces and state-dependent constraint sets. We show how the proposed methodology can be applied to complex inventory replenishment problems where analytical solutions are intractable. We also benchmark the proposed algorithm against state-of-the-art RL algorithms and commonly used replenishment heuristics and find that the proposed algorithm considerably outperforms existing methods by as much as 14.7\% on average in various supply chain settings. This improvement in performance of PARL over benchmark algorithms can be attributed to better inventory cost management, especially in inventory constrained settings. Furthermore, in a simpler back order setting where the optimal solution is tractable, we find that the RL based policy also converges to the optimal policy. Finally, to make RL algorithms more accessible for inventory management researchers, we also discuss a modular Python library developed that can be used to test the performance of RL algorithms with various supply chain structures. This library can spur future research in developing practical and near-optimal algorithms for inventory management problems., Comment: Prior shorter version accepted to NeurIPS 2021 Deep RL Workshop. Authors are listed in alphabetical order

We consider reinforcement learning (RL) in episodic Markov decision processes (MDPs) with linear function approximation under drifting environment. Specifically, both the reward and state transition functions can evolve over time but their total variations do not exceed a $\textit{variation budget}$. We first develop $\texttt{LSVI-UCB-Restart}$ algorithm, an optimistic modification of least-squares value iteration with periodic restart, and bound its dynamic regret when variation budgets are known. Then we propose a parameter-free algorithm $\texttt{Ada-LSVI-UCB-Restart}$ that extends to unknown variation budgets. We also derive the first minimax dynamic regret lower bound for nonstationary linear MDPs and as a byproduct establish a minimax regret lower bound for linear MDPs unsolved by Jin et al. (2020). Finally, we provide numerical experiments to demonstrate the effectiveness of our proposed algorithms.

We envision AI marketplaces to be platforms where consumers, with very less data for a target task, can obtain a relevant model by accessing many private data sources with vast number of data samples. One of the key challenges is to construct a training dataset that matches a target task without compromising on privacy of the data sources. To this end, we consider the following distributed data summarizataion problem. Given K private source datasets denoted by $[D_i]_{i\in [K]}$ and a small target validation set $D_v$, which may involve a considerable covariate shift with respect to the sources, compute a summary dataset $D_s\subseteq \bigcup_{i\in [K]} D_i$ such that its statistical distance from the validation dataset $D_v$ is minimized. We use the popular Maximum Mean Discrepancy as the measure of statistical distance. The non-private problem has received considerable attention in prior art, for example in prototype selection (Kim et al., NIPS 2016). Our work is the first to obtain strong differential privacy guarantees while ensuring the quality guarantees of the non-private version. We study this problem in a Parsimonious Curator Privacy Model, where a trusted curator coordinates the summarization process while minimizing the amount of private information accessed. Our central result is a novel protocol that (a) ensures the curator accesses at most $O(K^{\frac{1}{3}}|D_s| + |D_v|)$ points (b) has formal privacy guarantees on the leakage of information between the data owners and (c) closely matches the best known non-private greedy algorithm. Our protocol uses two hash functions, one inspired by the Rahimi-Recht random features method and the second leverages state of the art differential privacy mechanisms. We introduce a novel "noiseless" differentially private auctioning protocol for winner notification and demonstrate the efficacy of our protocol using real-world datasets., Comment: To appear in the 33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada

The reliability function of variable-rate Slepian-Wolf coding is linked to the reliability function of channel coding with constant composition codes, through which computable lower and upper bounds are derived. The bounds coincide at rates close to the Slepian-Wolf limit, yielding a complete characterization of the reliability function in that rate regime. It is shown that variable-rate Slepian-Wolf codes can significantly outperform fixed-rate Slepian-Wolf codes in terms of rate-error tradeoff. The reliability function of variable-rate Slepian-Wolf coding with rate below the Slepian-Wolf limit is determined. In sharp contrast with fixed-rate Slepian-Wolf codes for which the correct decoding probability decays to zero exponentially fast if the rate is below the Slepian-Wolf limit, the correct decoding probability of variable-rate Slepian-Wolf codes can be bounded away from zero., Comment: This is an old manuscript written in 2007-2008 based on our 2007 Allerton conference paper with the same title

This paper investigates compression of data encrypted with block ciphers, such as the Advanced Encryption Standard (AES). It is shown that such data can be feasibly compressed without knowledge of the secret key. Block ciphers operating in various chaining modes are considered and it is shown how compression can be achieved without compromising security of the encryption scheme. Further, it is shown that there exists a fundamental limitation to the practical compressibility of block ciphers when no chaining is used between blocks. Some performance results for practical code constructions used to compress binary sources are presented.

We consider Slepian-Wolf code design based on LDPC (low-density parity-check) coset codes for memoryless source-side information pairs. A density evolution formula, equipped with a concentration theorem, is derived for Slepian- Wolf coding based on LDPC coset codes. As a consequence, an intimate connection between Slepian-Wolf coding and channel coding is established. Specifically we show that, under density evolution, design of binary LDPC coset codes for Slepian-Wolf coding of an arbitrary memoryless source-side information pair reduces to design of binary LDPC codes for binary-input output-symmetric channels without loss of optimality. With this connection, many classic results in channel coding can be easily translated into the Slepian-Wolf setting., Comment: to appear at ISIT 2006

Disclosed are a method and system for video compression, wherein the video encoder has low computational complexity and high compression efficiency. The disclosed system comprises a video encoder and a video decoder, wherein the method for encoding includes the steps of converting a source frame into a space-frequency representation; estimating conditional statistics of at least one vector of space-frequency coefficients; estimating encoding rates based on the said conditional statistics; and applying Slepian-Wolf codes with the said computed encoding rates. The preferred method for decoding includes the steps of; generating a side-information vector of frequency coefficients based on previously decoded source data, encoder statistics, and previous reconstructions of the source frequency vector; and performing Slepian-Wolf decoding of at least one source frequency vector based on the generated side-information, the Slepian-Wolf code bits and the encoder statistics.

Three related innovations combine improved non-linear motion estimation, video coding, and video compression. The first system comprises a method in which side information is generated using an adaptive, non-linear motion model. This method enables extrapolating and interpolating a visual signal, including determining the first motion vector between the first pixel position in a first image to a second pixel position in a second image; determining a second motion vector between the second pixel position in the second image and a third pixel position in a third image; determining a third motion vector between the first pixel position in the first image and the second pixel position in the second image, the second pixel position in the second image, and the third pixel position in the third image using a non-linear model; and determining a position of the fourth pixel in a fourth image based upon the third motion vector. For the video compression element, the video encoder has low computational complexity and high compression efficiency. The disclosed system comprises a video encoder and a decoder. The encoder converts the source frame into a space-frequency representation, estimates the conditional statistics of at least one vector of space-frequency coefficients with similar frequencies, and is conditioned on previously encoded data. It estimates an encoding rate based on the conditional statistics and applies a Slepian-Wolf code with the computed encoding rate. The method for decoding includes generating a side-information vector of frequency coefficients based on previously decoded source data and encoder statistics and previous reconstructions of the source frequency vector. It also performs Slepian-Wolf decoding of a source frequency vector based on the generated side-information and the Slepian-Wolf code bits. The video coding element includes receiving a first reference frame having a first pixel value at a first pixel position, a second reference frame having a second pixel value at a second pixel position, and a third reference frame having a third pixel value at a third pixel position. It determines a first motion vector between the first pixel position and the second pixel position, a second motion vector between the second pixel position and the third pixel position, and a fourth pixel value for a fourth frame based upon a linear or nonlinear combination of the first pixel value, the second pixel value, and the third pixel value. A stationary filtering process determines the estimated pixel values. The parameters of the filter may be predetermined constants.