Speculative decoding is an effective method for lossless acceleration of large language models during inference. It uses a fast model to draft a block of tokens which are then verified in parallel by the target model, and provides a guarantee that the output is distributed identically to a sample from the target model. In prior works, draft verification is performed independently token-by-token. Surprisingly, we show that this approach is not optimal. We propose *Block Verification*, a simple draft verification algorithm that verifies the entire block jointly and provides additional wall-clock speedup. We prove that the proposed mechanism is optimal in the expected number of tokens produced each iteration and specifically is never worse than the standard token-level verification. Empirically, block verification provides modest but consistent wall-clock speedups over the standard token verification algorithm of 5\%-8\% in a range of tasks and datasets. Given that block verification does not increase code complexity, maintains the strong lossless guarantee of the standard speculative decoding verification algorithm, cannot deteriorate performance, and, in fact, consistently improves it, it can be used as a good default in speculative decoding implementations.

In this paper, we use a novel algorithmic approach to explore dialectal variation in American English speech. Without the need for human phonemic annotations, we are able to use an existing corpus transcribed in text form only. Our results show that, in general, American English dialects can be divided into two larger groups: dialects of the South (Texas to North Carolina except for peninsular Florida), and the rest of the country. Our results confirm some well-known results from dialectology, such as the pin-pen merger, but show that some other ones, such as the cot-caught merger, may be losing their isogloss boundaries. Moreover, we demonstrate that our algorithm can extend to dialectal features in other languages.

Classical multiple access channel (MAC) coding schemes assume independent transmitted signals, because the given messages are assumed to be independent and the remote transmitters cannot cooperate. However, when the messages are correlated, or in the extreme case of fully cooperative transmitters, a better performance may be achieved. We explore the performance gain of this collaboration and its connection to transmission synchronization. For totally asynchronous transmitters, we prove that transmitters' collaboration has no performance gain. This result is shown for finite-alphabet memoryless MACs and for the additive white Gaussian noise MAC.

This work discusses the source broadcasting problem, i.e. transmitting a source to many receivers via a broadcast channel. The optimal rate-distortion region for this problem is unknown. The separation approach divides the problem into two complementary problems: source successive refinement and broadcast channel transmission. We provide bounds on the loss incorporated by applying time-sharing and separation in source broadcasting. If the broadcast channel is degraded, it turns out that separation-based time-sharing achieves at least a factor of the joint source-channel optimal rate, and this factor has a positive limit even if the number of receivers increases to infinity. For the AWGN broadcast channel a better bound is introduced, implying that all achievable joint source-channel schemes have a rate within one bit of the separation-based achievable rate region for two receivers, or within log2T bits for T receivers.

Classical MAC coding schemes assume independent transmitted signals, because the given messages are assumed to be independent and the remote transmitters cannot cooperate. However, when the messages are correlated, or in the extreme case of fully cooperative transmitters, a better performance may be achieved. We explore the performance gain of this collaboration and its connection to transmission synchronization. For totally asynchronous transmitters we prove that transmitters collaboration has no performance gain. This result is proven for discrete-value channels. For continuous-value channels we focus on the collaboration gain of the AWGN MAC, indicating that it is bounded by a single bit and a factor of two. We suggest that this collaboration gain diminishes without synchronization as well.