EMA
EMA stands for exponential moving average, a way of smoothing a value over time by blending each new observation with the running average instead of replacing it. In Bittensor, the EMA documentation describes this smoothing as how validator-miner bonds are updated across epochs rather than recomputed from scratch each time.
The blend is controlled by a smoothing factor, often called alpha. A higher alpha weights the newest observation more, so the average reacts faster to change, while a lower alpha leans on accumulated history, so the average moves more slowly and smooths short-term noise (EMA).
How the EMA Update Works
Each epoch the EMA does not replace the old value—it blends. The updated bond is the new observation weighted by alpha plus the previous bond weighted by one minus alpha, so prior history is carried forward and only partly displaced by the latest epoch (EMA, Glossary: Validator-Miner Bonds).
That single coefficient sets the balance. An alpha near one makes the average track the newest observation closely; an alpha near zero leans on accumulated history and changes slowly, damping short-term noise (EMA).
Why Bonds Are Smoothed This Way
Bittensor applies this smoothing to validator-miner bonds so the bond state evolves gradually across epochs instead of resetting each time. A validator’s accumulated relationship to a miner builds and fades over many epochs rather than swinging with a single epoch’s result (EMA, Glossary: Validator-Miner Bonds).
The smoothed bond state then feeds consensus: Yuma Consensus reads it as input when it applies bonding at the epoch boundary, so the EMA shapes the bond history that settlement consumes rather than the settlement itself (Yuma Consensus, Emission).
Distinction from Liquid Alpha
The EMA is the smoothing method; liquid alpha is a feature that varies the EMA’s alpha coefficient by how well a validator aligns with consensus, instead of holding it fixed. The alpha low and alpha high bounds set the range that coefficient may take (Consensus-based Weights, Subnet Hyperparameters).
- EMA — the moving-average technique that blends new and prior values.
- Liquid alpha — varies the EMA coefficient by consensus alignment.
Distinction from Validator-Miner Bonds
Validator-miner bonds are the values being smoothed; the EMA is the rule for updating them. One is the quantity, the other is the method applied to it across epochs (Yuma Consensus).
- Validator-miner bonds — the accumulated validator-to-miner relationship.
- EMA — how those bonds are updated each epoch.
Distinction from a Simple Moving Average
An exponential moving average weights recent observations more heavily and lets older values fade gradually. A simple moving average takes the equal-weighted mean of a fixed recent window (EMA).
- EMA — recency-weighted blend where older values fade gradually.
- Simple moving average — equal-weighted mean of a fixed recent window.
Bond smoothing favors gradual accumulation over abrupt window resets (Glossary: Validator-Miner Bonds).
Distinction from Bonds Moving Average
The EMA is the smoothing technique applied to validator-miner bonds. Bonds moving average is the
per-subnet BondsMovingAverage hyperparameter that configures that smoothing on a selected netuid
(Subnet Hyperparameters,
Yuma Consensus).
- EMA — the smoothing technique applied to bonds.
- Bonds moving average — the per-subnet hyperparameter that configures that smoothing.
Distinction from Yuma Consensus
EMA (exponential moving average) names the smoothing technique that blends each epoch’s validator-miner bond update with prior bond history. It describes how bond state evolves over time inside the bond path, not the settlement step that converts weight submissions into emission shares (EMA, Glossary: Validator-Miner Bonds).
Yuma Consensus is the on-chain mechanism that runs at the epoch boundary. It reads the weight matrix, applies clipping and bonding (using EMA-smoothed bond state as input), and converts the filtered result into miner incentives and validator dividend shares (Yuma Consensus, Emission).
The bonds moving average section above names the hyperparameter that configures smoothing magnitude on a subnet. EMA is the technique; Yuma is the full settlement pass that credits outcomes after weights and bond rules run—not the smoothing formula by itself (Subnet Hyperparameters).
Readers should treat EMA as bond-path smoothing vocabulary, and Yuma as the epoch-boundary mechanism that allocates rewards from consensus.
- EMA — exponential moving average technique for validator-miner bond updates.
- Yuma Consensus — epoch-boundary settlement from filtered weights to emission shares.
Distinction from Alpha Sigmoid Steepness
The alpha that the steepness curve assigns is the EMA smoothing coefficient used when updating validator-miner bonds. EMA is the technique that blends each epoch’s new bond value with the prior one, while alpha sigmoid steepness shapes how the coefficient feeding that EMA is set from a validator’s consensus alignment under the liquid alpha feature (EMA, Consensus-based Weights, Glossary: Validator-Miner Bonds).
- EMA — the moving-average technique that blends new and prior bond values.
- Alpha sigmoid steepness — shapes the coefficient that EMA uses under liquid alpha.
Reader Boundary
EMA should not be read as a subnet hyperparameter value or as liquid alpha itself. It names the exponential moving-average technique Bittensor uses to update validator-miner bonds across epochs (EMA, Yuma Consensus).
Live bond-smoothing behavior on a subnet also depends on that netuid’s hyperparameter state, including bonds moving average and any liquid-alpha settings (Subnet Hyperparameters).