It is common to reject undesired outputs of Large Language Models (LLMs); however, current methods to do so require an excessive amount of computation to re-sample after a rejection, or distort the distribution of outputs by constraining the output to highly improbable tokens. We present a method, Approximately Aligned Decoding (AprAD), to balance the distortion of the output distribution with computational efficiency, inspired by algorithms from the speculative decoding literature. AprAD allows for the generation of long sequences of text with difficult-to-satisfy constraints, while amplifying low probability outputs much less compared to existing methods. We show through a series of experiments that the task-specific performance of AprAD is comparable to methods that do not distort the output distribution, while being much more computationally efficient.

While instruction-tuned language models have demonstrated impressive zero-shot generalization, these models often struggle to generate accurate responses when faced with instructions that fall outside their training set. This paper presents Instructive Decoding (ID), a simple yet effective approach that augments the efficacy of instruction-tuned models. Specifically, ID adjusts the logits for next-token prediction in a contrastive manner, utilizing predictions generated from a manipulated version of the original instruction, referred to as a noisy instruction. This noisy instruction aims to elicit responses that could diverge from the intended instruction yet remain plausible. We conduct experiments across a spectrum of such noisy instructions, ranging from those that insert semantic noise via random words to others like 'opposite' that elicit the deviated responses. Our approach achieves considerable performance gains across various instruction-tuned models and tasks without necessitating any additional parameter updates. Notably, utilizing 'opposite' as the noisy instruction in ID, which exhibits the maximum divergence from the original instruction, consistently produces the most significant performance gains across multiple models and tasks.

To better understand the source of the bias, in the first part of this post, we have explored the example of a firm that is interested in testing the effectiveness of an ad campaign. The firm has information on its current ad spending and on the level of sales. The problem arises because the firm is uncertain on whether it should condition its analysis on the level of past sales. I import the data generating process dgp_pretest() from src.dgp and some plotting functions and libraries from src.utils. We have data on 1000 different markets, for which we observe current sales, the amount spent in advertisement and past sales.

Japanese astronomers have developed a new artificial intelligence (AI) technique to remove noise in astronomical data due to random variations in galaxy shapes. After extensive training and testing on large mock data created by supercomputer simulations, they then applied this new tool to actual data from Japan's Subaru Telescope and found that the mass distribution derived from using this method is consistent with the currently accepted models of the Universe. This is a powerful new tool for analyzing big data from current and planned astronomy surveys. Wide area survey data can be used to study the large-scale structure of the Universe through measurements of gravitational lensing patterns. In gravitational lensing, the gravity of a foreground object, like a cluster of galaxies, can distort the image of a background object, such as a more distant galaxy.

Guided ultrasonic wave localization uses spatially distributed multistatic sensor arrays and generalized beamforming strategies to detect and locate damage across a structure. The propagation channel is often very complex. Methods can compare data with models of wave propagation to locate damage. Yet, environmental uncertainty (e.g., temperature or stress variations) often degrade accuracies. This paper uses an uncertainty-aware deep neural network framework to learn robust localization models and represent uncertainty. We use mixture density networks to generate damage location distributions based on training data uncertainty. This is in contrast with most localization methods, which output point estimates. We compare our approach with matched field processing (MFP), a generalized beamforming framework. The proposed approach achieves a localization error of 0.0625 m as compared to 0.1425 m with MFP when data has environmental uncertainty and noise. We also show that the predictive uncertainty scales as environmental uncertainty increases to provide a statistically meaningful metric for assessing localization accuracy.

Amazon's latest low-cost Alexa-powered smart speaker, the third-generation Echo Dot, looks better, sounds better, but still costs the same budget-friendly £50. When the second-generation Echo Dot launched in the UK it had very little in the way of competition. Having everything that was good about Amazon's Alexa voice assistant squeezed into a smaller package half the price of the bigger Echo, it was a no-brainer. But now the Echo Dot has some stiff competition from Google, in the form of the £50 Home Mini and its Assistant, and Amazon's own second-generation Echo, which is now only £90. The Echo Dot is Amazon's small, puck-like smart speaker with the firm's Alexa voice assistant built in. New for the third generation is a softer, more rounded aesthetic with fabric sides.

How do you have to distort a face so that facial recognition algorithms no longer see a face–and evade the technology that has become so pervasive in our world? That was the question the Seoul-based artistic duo Shin Seung Back and Kim Yong Hun posed to a group of 10 different painters. The result is their series Nonfacial Portrait, a striking collection of painted portraits that evade the algorithms. The paintings, which are currently on display at the Seoul Museum of Art, are each so wildly different that you wouldn't know they were all inspired by a single photo of Yong Hun. One has a sky blue outline of a bust, with the eyes, mouth, and nose scattered around the canvas.

During my past two decades working in the IT field, I've seen artificial intelligence technologies move from conceptual to practical -- with machine learning techniques at the forefront, becoming more accessible, even for teams without specialized expertise. With increased use of predictive modeling across a wide variety of teams, it's critical for leaders and managers to be aware of common issues that can distort the results of their teams' work. Here are nine common pitfalls to avoid, and best practices to follow, for a reliable machine learning process. The starting point of any machine learning program is to select the training data. Typically, organizations have some data available or can identify relevant external suppliers, such as government entities or industry associations.